Forecast Pro Unlimited V6 User's Guide

7/27/2019 Forecast Pro Unlimited V6 User's Guide

http://slidepdf.com/reader/full/forecast-pro-unlimited-v6-users-guide 1/193



Copyright Notice

Copyright © 1990 - 2009 by Business Forecast Systems, Inc. All RightsReserved Worldwide. No part of this manual may be reproduced, transmitted,transcribed, stored in a retrieval system, or translated into any human or

computer language, in any form or by any means, electronic, mechanical,magnetic, optical, chemical, manual or otherwise, without express written

permission of Business Forecast Systems, Inc., 68 Leonard Street, Belmont,MA 02478 USA.

Disclaimer

Business Forecast Systems, Inc. makes no representations or warranties withrespect to the contents hereof and specifically disclaims any implied warranties of merchantability or fitness for any particular purpose. Further,Business Forecast Systems, Inc. reserves the right to revise this publication

and to make changes from time to time in the contents hereof withoutobligation of Business Forecast Systems, Inc. to notify any person or organization of such revision or change.

Trademarks

Forecast Pro is a registered trademark of Business Forecast Systems, Inc.

Other product names mentioned in this manual are trademarks or registered trademarks of their respective companies and are hereby acknowledged.



ii i

Contents

GETTING STARTED Before You Begin ............................................................................................ 3 Forecasters and Collaborators ........................................................................... 3 How to Use This Manual .................................................................................. 4 Statistical Reference Manual (PDF only) ......................................................... 6 Checking Your Forecast Pro Unlimited Package ............................................. 6 What You Need to Run Forecast Pro Unlimited .............................................. 7 Registering Your Forecast Pro Unlimited Package .......................................... 7 Future Development of Forecast Pro Unlimited ............................................... 7 Installing Forecast Pro Unlimited ................................................................. 9 Installation Instructions..................................................................................... 9 A Quick Overview of Forecasting ............................................................... 11 What is Statistical Forecasting? ...................................................................... 11 Forecasting Methodologies ............................................................................. 13 Some Forecasting Tips .................................................................................... 15



iv

TUTORIAL The Basics ...................................................................................................... 19 The Main Menu .............................................................................................. 19 An Overview ................................................................................................... 21 Setting the Defaults ......................................................................................... 21 Defining the Script .......................................................................................... 23 Reading in the Data ......................................................................................... 24 Generating the Forecasts ................................................................................. 26 Viewing the Forecast Report .......................................................................... 27 Customizing the Graph View.......................................................................... 29 Overriding the Forecasts ................................................................................. 31 Working With a Hot List ................................................................................ 33 Saving the Forecasts ....................................................................................... 35 Saving the Project ........................................................................................... 38 Using Forecast Modifiers ............................................................................. 39 Specifying Forecast Modifiers for Individual Items ....................................... 39 Specifying Forecast Modifiers for Multiple Items ......................................... 41 Specifying Custom Forecast Modifiers .......................................................... 42 Collaborating With Colleagues .................................................................... 43 Sharing a Stand-alone Project ......................................................................... 43 Updating a Forecast Project ........................................................................ 47 Defining the Forecasting Process ................................................................... 47 Step One: Updating Your Data Files .............................................................. 48 Step Two: Opening Your Forecast Project ..................................................... 48 Step Three: Saving Your Work....................................................................... 50 Building Event Models ................................................................................. 51 Modeling Promotions ..................................................................................... 52 Modeling Weekly Data ................................................................................... 55 Adjustment for Outliers .................................................................................. 59 More Adjustments for Promotions.................................................................. 60 Building Multiple-Level Models .................................................................. 61 Bottom-up and Top-down Forecasting ........................................................... 61 Defining Groups Using Attribute Fields ......................................................... 64



v

Combining Multiple Level and Event Adjustment Models ............................ 65 Short-Lived Products ...................................................................................... 66 Using Weighting Transformations .............................................................. 69 Adjusting for 4 vs. 5 Week Periods ................................................................ 69 Other Uses for Weighting Transformations .................................................... 72 Detecting and Correcting Outliers .............................................................. 73 How Outlier Detection and Correction Works ............................................... 74 Reviewing the Outliers ................................................................................... 75 Correcting the Outliers.................................................................................... 76 Out-of-sample Testing .................................................................................. 79 Setting Up the Script for Evaluation ............................................................... 80 Examining the Analytic Output ...................................................................... 81 Comparing an Alternative Model ................................................................... 82 Operating From the Command Line .......................................................... 85 Command-Line Operation .............................................................................. 85

REFERENCE Setting Up Your Data ................................................................................... 91 Overview ......................................................................................................... 91 Selecting a Data Format .................................................................................. 95 Spreadsheet (XLS, XLSX and WK*) Formats ............................................... 96 Text (MLT) Formats ..................................................................................... 101 Using ODBC ................................................................................................. 103 Using Attribute Fields to Define a Hierarchy ............................................... 106 Saving Your Work ...................................................................................... 109 Forecast Projects ........................................................................................... 110 Numeric Output Files .................................................................................... 111 Formatted Forecast Reports .......................................................................... 112 Override Reports ........................................................................................... 113 Outlier Reports .............................................................................................. 113 Working with Scripts and Hierarchies ..................................................... 115 Defining the Script ........................................................................................ 115



vi

Defining the Hierarchy ................................................................................. 116 Reconciling the Hierarchy ............................................................................ 118 Using Forecast Modifiers ........................................................................... 123 Model Specification Modifiers ..................................................................... 123 Event Model Modifiers ................................................................................. 128 The Weighting Transformation Modifier ..................................................... 129 Outlier Detection/Correction Modifiers........................................................ 129 Reconciliation Modifiers .............................................................................. 130 Miscellaneous Custom Modifiers ................................................................. 131 Command Reference .................................................................................. 133 Operations Overview .................................................................................... 133 The Forecast Pro Unlimited Interface ........................................................... 134 The Forecast Report View ............................................................................ 140 The Graph View ............................................................................................ 141 The Override Forecasts View ....................................................................... 144 The Override Report View ........................................................................... 147 The Outlier Report View .............................................................................. 147 The Numeric Output View............................................................................ 147 Menu Items and Dialog Boxes ...................................................................... 151 Command-line Operation.............................................................................. 164 Understanding How Forecast Overrides are Allocated .......................... 167 Adjustments Made to a Single Level ............................................................ 167 Adjustments Made to More Than One Level ............................................... 169 Examples Where Overrides Cannot be Fully Allocated ............................... 171 Summation .................................................................................................... 173 Appendixes................................................................................................... 175 Speeding Up the Program ............................................................................. 175 End-User License Agreement ....................................................................... 177 Suggested Reading ........................................................................................ 181 Index ............................................................................................................. 183



1

GETTING STARTED



3

Chapter 1

Before You Begin

Forecasters and Collaborators

Forecast Pro Unlimited is available in two different types of licenses—

Forecaster licenses and Collaborator licenses. The type of license you install

will dictate the functionality you can access. A Forecaster has access to all of

the functionality described in this manual. A Collaborator can view and

manipulate a forecast project created by a Forecaster (including viewing

graphs and reports, adding overrides and comments, saving output, etc.) but

cannot generate new statistical forecasts.

Operationally, the two installations are very similar, however, a Collaborator does not have access to any of the functionality relating to reading historical

data and generating statistical forecasts. Thus, there are sections of this

manual which will not be relevant to Collaborators.



4 Before You Begin

How to Use This Manual

This manual is divided into three parts.

Part 1 Getting Started

Chapter 1 describes the Forecast Pro Unlimited system requirements and product support policy.

Chapter 2 describes the installation procedure.

Chapter 3 provides an overview of forecasting.

Part 2 Tutorial

Lesson 1 walks you through a sample forecasting session using automaticmodel selection.

Lesson 2 teaches you how to use forecast modifiers to dictate model selection.

Lesson 3 teaches you how to collaborate with colleagues to establish the finalforecast.

Lesson 4 teaches you how to update your forecast project as new data becomeavailable.

Lesson 5 teaches you how to build event models to capture promotionaleffects.

Lesson 6 teaches you how to set up and forecast a multiple-level script.

Lesson 7 teaches you how to use weights to adjust for trading day impacts or to forecast new products based on similar products.

Lesson 8 teaches you how to use Forecast Pro Unlimited’s outlier detectionand correction functionality.

Lesson 9 teaches you how to assess forecasting performance using a holdoutsample approach.



Statistical Reference Manual (PDF only) 5

Lesson 10 teaches you how to operate the program from the command line.

All of the lessons use sample data provided with the software.

Part 3 Reference

Chapter 1 explains how to select the appropriate data format and how to set

up your database.

Chapter 2 describes the various reports and output files that can be saved.

Chapter 3 describes how to set up a Forecast Pro Unlimited script and alsodiscusses the strategies that Forecast Pro Unlimited uses to produce consistentforecasts for hierarchical data.

Chapter 4 describes all available forecast modifiers and their functions.

Chapter 5 is a complete reference to Forecast Pro Unlimited menus, optionsand commands.

Chapter 6 describes how Forecast Pro Unlimited reconciles forecastadjustments between different levels of the forecasting hierarchy.

The Appendixes describe how to speed up processing and provide other supplementary information.

Many users won't want to work their way through each and every part of thismanual. However, we urge you to go through at least Chapters 1-3 of GettingStarted, Lesson 1, 2 3 & 4 of the Tutorial and as many other lessons as youwant. This will give you a good overview of Forecast Pro Unlimited

operations and capabilities.

When you are ready to set up your own data, please consult Chapter 1 of Part

3 Reference.



6 Before You Begin

Statistical Reference Manual (PDFonly)

The Forecast Pro Statistical Reference Manual describes the statistical

techniques, statistics, and strategies that are implemented in Forecast Pro. It is

not necessary that you fully understand, or even read, this manual in order to

produce accurate forecasts with the product.

The manual is provided in electronic format (pdf). When Forecast Pro

Unlimited is installed, the file is copied into the program directory. You can

access the file directly or via the Help menu in Forecast Pro Unlimited.

Chapter 1 is a detailed statistical reference. It explains the theory behind Forecast Pro Unlimited’s statistical models, diagnostics and methodology.

Chapter 2 is a discussion of some of the general considerations that apply toautomatic batch forecasting.

Checking Your Forecast ProUnlimited Package

Your Forecast Pro Unlimited package should contain the following items:

This User's Manual. This book walks you through the installation and use of Forecast Pro Unlimited.

One installation CD.

A warranty registration card.

A maintenance and support contract.

If your package is missing any of the above items please contact Business

Forecast Systems, Inc.



What You Need to Run Forecast Pro Unlimited 7

What You Need to Run ForecastPro Unlimited

A computer running a Windows operating system.

A minimum of 512 MB of random access memory (RAM). Additional

memory will greatly enhance program performance. 4 gigabytes isrecommended.

A hard disk drive with 30 MB of free space.

Registering Your Forecast ProUnlimited Package

Please take a moment NOW to fill out and mail the Registration Card youreceived with Forecast Pro Unlimited or visit www.forecastpro.com and

register on-line. Registering your software entitles you to the following benefits:

• Free maintenance and support service for one year. This service provides program updates and unlimited technical support for the firstyear you license the program. After the first year, this service isavailable on a subscription basis. Technical support is provided via thetelephone (during regular business hours Eastern Standard Time), viaemail, mail and fax.

• Automatic notification of upgrades, revisions and new products.

• Special pricing on upgrades, revisions and new products.

Future Development of ForecastPro Unlimited

BFS has made many changes to Forecast Pro Unlimited to accommodatespecial needs and circumstances in the corporate environment. BFS willcontinue to improve Forecast Pro Unlimited in the years to come. If you



8 Before You Begin

encounter problems or have any suggestions for improvements or additionalfeatures please contact the BFS product development staff.



9

Chapter 2

Installing Forecast Pro Unlimited

Installation Instructions

To install Forecast Pro Unlimited on your hard disk perform the followingsteps.

If you have an Internet connection, make sure that it is active.

Place the Forecast Pro Unlimited CD in your CD drive (we’ll use “D:” for these instructions).

Select Run from the Start menu.

Enter D:\SETUP for the command line and click OK.

Follow the directions that appear on the screen.

Setup will then create the program directory, data directories, copy theappropriate files, create a launch icon on the program menu and place alaunch icon on the desktop.



10 Installing Forecast Pro Unlimited

If you have any difficulty installing Forecast Pro Unlimited please do nothesitate to contact BFS technical support (phone: 617 484-5050, email:[email protected]).

If you wish to remove Forecast Pro Unlimited from your computer, use theAdd/Remove Programs utility in the Windows Control Panel.



11

Chapter 3

A Quick Overview of Forecasting

What is Statistical Forecasting?

Everybody forecasts, whether they know it or not. Businesses have to forecastfuture events in order to plan production, schedule their work force, or prepareeven the simplest business plan.

Most business forecasting is still judgmental and intuitive. Sometimes this isappropriate. People must integrate information from a large variety of

sources—qualitative and quantitative—and this is probably best done byusing the extraordinary pattern recognition capabilities of the human brain.Unfortunately, many companies also use judgmental forecasting where they

should not.

Not everyone understands the concept of forecasting. It tends to get mixed upwith goal setting. If a company asks its salespeople to forecast sales for their territories, these “forecasts” often become the yardsticks by which they are judged.

The main advantage of statistical forecasting is that it separates the process of forecasting from that of goal setting, and makes it systematic and objective.



12 A Quick Overview of Forecasting

Objective, quantitative forecasting can help almost any business substantially.There is, in other words, value added for business.

The future is uncertain, and this uncertainty must be represented quantitatively. Statistical forecasting represents uncertainty as a probability

distribution. Two kinds of information are needed to describe the distribution:the point forecasts and the confidence limits.

A point forecast is the mean value of the distribution of future values, and can be thought of as a best estimate of the future value. Its upper and lower confidence limits describe the spread of the distribution above and below the point forecast. Forecast Pro Unlimited depicts this information graphically aswell as numerically.

The upper confidence limit is often calibrated to the ninety-fifth percentile.This means that the actual value should fall at or below the upper confidencelimit about 95% of the time. You can set the percentiles of both the upper and lower confidence limits.

Let's illustrate this idea with an example. Suppose you were in charge of forecasting widget sales for your company. If you wanted to determineexpected revenues for next month, you would be most interested in the pointforecast, since it is the mean value of the distribution. The point forecast givesyou the minimum expected forecast error.



Forecasting Methodologies 13

On the other hand, suppose you wanted to know how many widgets to produce. If you overproduce, warehousing costs will be excessive. But if youunderproduce, you will probably lose sales. Since the cost of lost sales isusually greater than the cost of overstocking, you will be most interested inthe upper confidence limit. The upper confidence limit tells you how manywidgets to produce to limit the chance of “stocking out” to less than 5%.

Forecasting Methodologies

A wide variety of statistical forecasting techniques are available, ranging fromvery simple to very sophisticated. All of them try to capture the statisticaldistribution that we have just discussed.

Forecast Pro Unlimited offers the five methodologies that are mostappropriate for automated business forecasting: simple moving averages,discrete data models (Poisson or negative binomial), Croston’s intermittentdata model, exponential smoothing and Box-Jenkins. Extended exponential

smoothing models (“event models”) are also included to accommodate promotions and weekly seasonality. All of these models are univariate techniques. They forecast the future entirely from statistical patterns in the past.

Thus you must have historic records, preferably for several years, of thevariable you want to forecast.

Forecast accuracy depends upon the degree to which statistical data patternsexist, and their constancy over time. The more regular the series, the moreaccurate the forecasts.

The simple moving average is widely used in business, mostly because it is soeasy to implement. However, it is really only appropriate for very short or very irregular data sets, where statistical features like trend and seasonalitycannot be meaningfully determined.

Discrete data models are used for data consisting of small whole numbers.These models are characteristically used to model a slow-moving item for which most orders are for only one piece at a time. Forecasts are nontrended and nonseasonal.



14 A Quick Overview of Forecasting

Croston’s intermittent demand model is not a widely known or used technique but, in certain circumstances, it is extremely useful. It is usually used to modeldata in which a significant number of periods have zero demand but the non-zero orders may be substantial. This is characteristic of a slow-moving itemwhich is ordered to restock a downstream inventory. Forecasts are nontrended and nonseasonal.

Exponential smoothing models are widely applicable. They are also widelyused, because of their simplicity, accuracy, and ease of use. Their robustnessmakes them ideal even when the data are short and/or volatile. Exponentialsmoothing works by identifying and extracting trend and seasonality, and extrapolating them forward.

Box-Jenkins is a more elaborate statistical method than exponentialsmoothing. Box-Jenkins works by capturing the historic correlations of thedata, and extrapolating them forward. It often outperforms exponentialsmoothing in cases when the data are fairly long and nonvolatile. However, itdoesn't usually perform as well when the data are statistically messy.

You can use Forecast Pro Unlimited’s expert selection to automaticallychoose the appropriate forecasting technique for each item forecasted.Alternatively, you can dictate that a specific method be used. If you arealready familiar with statistical forecasting, you can use Forecast ProUnlimited to customize your models. It provides extensive diagnostics and statistical tests to help you make informed decisions.

If your data are driven by promotions or exhibit hard-to-capture seasonality(e.g., weekly data) you may want to experiment with event models. Thesemodels allow you to assign each period into logical categories and incorporate

an adjustment for each category. For example if you establish a category for promoted months then your model would include an adjustment for promoted months. If you ran three different types of promotions you could establishthree categories and have a different adjustment for each type of promotion.To build a monthly seasonal model for weekly data you would establishtwelve categories based upon which month the week was in.

If you are new to forecasting and these techniques seem a little intimidating,don't worry. We designed Forecast Pro Unlimited to guide you completely



Some Forecasting Tips 15

through the forecasting process. Just follow the program's advice all the wayto accurate forecasts.

Some Forecasting Tips

Forecast Pro Unlimited uses the history of your data to forecast the future.

Thus it is extremely important that your data be as accurate and as completeas possible. Keep in mind the rule, “Garbage in, garbage out!”

You will also want to give some thought to what data you should forecast. If

you want to forecast demand for your product you should probably input and forecast incoming orders rather than shipments, which are subject to production delays, warehousing effects, labor scheduling, etc. Better yet, use point-of-sale data. Many corporations are making large investments to obtaindata as close to true demand as possible.

The more data you can supply the program the better. The program can work

with as few as five data points, but the forecasts from very short series aresimplistic. Although collecting additional data may require some effort, it isusually worth it.

If your data are seasonal, it is particularly important that you have adequatedata length. The automatic model selection algorithms in Forecast ProUnlimited will not consider seasonal models unless you have at least twoyears worth of data. This is because you need at least two samples for eachmonth or quarter to distinguish seasonality from one-time irregular patterns.Ideally you should use three or more years of data to build a seasonal model.

Finally, remember that forecasts are never perfect. Forecast Pro Unlimited bases its forecasts solely on the past history of your data. If you knowsomething that Forecast Pro Unlimited did not, then you may want to adjustthe forecasts judgmentally. For instance, you may know of future events like alarge upcoming sale or the introduction of a new product. You can use thequantitative forecasts as a starting point, and apply your own insight and knowledge of future events to improve them.



2

TUTORIAL



19

Lesson 1

The Basics

Before you can use Forecast Pro Unlimited effectively, you must learn some basic concepts. This tutorial will guide you through these fundamentals. Thetutorial is divided into staged lessons. You should complete them all to get asound introduction to Forecast Pro Unlimited.

This first lesson provides an overview of Forecast Pro Unlimited and guidesyou through a sample forecasting session.

Before you can use Forecast Pro Unlimited, you must install it. If you havenot already done so, follow the installation instructions discussed in Chapter 2.

The Main Menu

To start Forecast Pro Unlimited click the Start button, select All Programs and click the Forecast Pro Unlimited icon. After the program is loaded, you will

see a display like the one below.



20 The Basics

menu bar Æ toolbar Æ

dialog bar Æ

Navigator Æ

Hot ListÆ

status bar Æ

← ForecastReportwindow

The main window, entitled Forecast Pro Unlimited consists of a menu bar , atoolbar, a dialog bar, a Navigator (currently unavailable), a Hot List area, aForecast Report window and a status bar.

The menu bar provides access to all procedures. The most commonly used procedures can also be accessed instantly via the icons on the toolbar. Thedialog bar is currently blank. After data have been read in, the dialog bar willinclude the span of the data, the specified holdout sample and the currentlydisplayed units. The Navigator and Hot List are currently unavailable. After data have been read in, the Navigator and Hot List can be used to selectindividual items that you wish to view.

The Forecast Report window is used to display the forecasting model, within-sample statistics, historic data and forecasts. The Forecast Report window isone of three context-sensitive views. A graph view will become available after data has been read in. An override view will become available after forecastshave been generated. Active views display information for the item currentlyselected on the Navigator. When more than one view is active, the programwill tile them. The views are toggled on or off using the green icons or theView menu.

Three additional noncontext-sensitive views are also available—the overridereport view, the outlier report view and the numeric output view. These viewsdisplay information for all items forecasted and are toggled on or off using theyellow icons or the View menu.



An Overview 21

An Overview

Running Forecast Pro Unlimited consists of executing the following fivesteps:

1. Prepare the database. This is performed outside of the program and entails selecting the type of data format to use and preparing the files.Creating your data files is discussed in the first chapter of theReference section. For this lesson you will use sample data files provided with the package.

2. Specify the data format and default options. This is accomplished byselecting the appropriate entries from the Options dialog box. Alloptions set by the user will be used for the current forecast project.You can also save your selections as the default for future projects.

3. Create the script. The script is a list of the data files to be forecasted and any special instructions. It is created using the Define Script

dialog box.

4. Select Operations>Read Data. Forecast Pro Unlimited will then read the data from disk into RAM and display the starting and ending datesfor the forecast run on the dialog bar.

5. Select Operations>Forecast. The software will then execute the

current script to prepare the forecasts. Once the forecasts have beencreated you may view them graphically, adjust them and save them todisk.

The remainder of this lesson will guide you through steps 2-5.

Setting the Defaults

To set the default options click the teal Options icon ( ) which will open the

Options dialog box shown below.



22 The Basics

There are six pages of options on the dialog box accessible via the tabs. Themore commonly changed options are found on the Basic tab shown above.

The Formats tab is used to set details of the data file interface. The defaultshave been carefully chosen and should only need to be changed for usersusing comma-delimited data files, or who do not follow the US conventionsfor the decimal separator. The Outliers tab allows you to control the settingsfor outlier detection and correction. This will be covered in Lesson 8. TheOverrides tab allows you to customize the override facility. The Command Line tab allows you to control the settings for command-line operation. Thiswill be covered in Lesson 10. The Advanced Controls tab is used to set details

of the optimization routines.

Let’s take a closer look at the Basic tab. Forecast Pro Unlimited reads in thehistorical data from the Input Data directory. Project files are saved to theProjects directory. Forecast files and reports are written to the Outputdirectory. The Data Mode specifies the format of the historical data. (Theforecast output can be saved in any supported format.)

You can set the percentiles for the confidence limits, the lead time for safetystock calculations, the length of the forecasts (forecast horizon) and can tellForecast Pro Unlimited whether or not to allow negative forecasts.

The data for this lesson were copied into your Forecast Pro Unlimited inputfolder when you installed the program (normally Forecast Pro Unlimited\Inputin your Shared Documents folder). Make sure that the Input Data directory isset correctly and that the other options match those shown above. Click theOK button to accept the defaults and return to the main menu.

If you are using a Forecaster license: proceed to the next section—Definingthe Script.



Defining the Script 23

If you are using a Collaborator license: the next three sections (Defining theScript, Reading in the Data and Generating the Forecasts) describe operationsthat are not supported. The first step in a Collaborator session is to open aforecast project that was created by someone using a Forecaster license. Select

File>Open and open the forecast project “Tutorial 1” then skip the next threesections and proceed to the Viewing the Forecast Report section.

Defining the Script

The next step is to define the script. The script lists the data file(s) you wish toforecast.

In this example we will prepare sales forecasts for 123 Bakery. The dataconsist of monthly sales of Muffins and Cakes.

Click the red Script icon ( ) to call up the Define Script dialog box shown

below.

You can type the filenames on the script, one to a line, or you can double-click the filename displayed in the list box to insert the filename onto the

script. If you are in ODBC mode, the tables and queries in your databasereplace filenames.

Double-click the filename 123 Bakery to insert it on the first line of the script.When you list a data file on a line of the script the program will forecast every

series in the data file, with the exception of helper variables, whose namesmust begin or end with an underbar . Thus, our first line instructs the programto forecast all series found in 123 Bakery.xls.



24 The Basics

Your script should look like the one above. Accept the script and return to themain menu by clicking the OK button.

Reading in the Data

You may have noticed that the red Read Data icon ( ) which had been

grayed out is now activated. Click this icon to read the data into memory.When you read data, any existing data in memory is replaced.

Click the green View Graph icon ( ) to open the graph view and display agraph for 123 Bakery. Double click Total on the Navigator to expand the Navigator’s tree. You’ll see that the total sales breakdown into Muffins and Cakes. Select Total>Muffins on the Navigator to display its graph. If youexpand Total>Muffins on the Navigator you’ll discover a customer-level breakdown and finally an SKU level. Experiment with the Navigator until youare comfortable with its operation. In the next lesson, we’ll show you someshortcuts for navigating through complex hierarchies.

After you have explored the data, arrange your display so that it matches theone shown below.



Reading in the Data 25

You probably noticed that after the data had been read in the dialog bar wasupdated to include the time span of the data and the holdout sample.

The dialog bar displays the earliest and latest period found for the entirescript. The ending date is very important. If an individual item has data on theending date, the item is active and will be forecasted. If an item’s history ends prior to the ending date, the item is inactive and will not be forecasted. If youdo not adjust the starting or ending dates Forecast Pro Unlimited will prepareforecasts using all available data for each active series.

For our current example we will not adjust the time span of the data nor willwe define a holdout sample.



26 The Basics

Generating the Forecasts

Click the red Forecast icon ( ) to generate the forecasts.

Notice that the Graph view now includes forecast components and theForecast Report view includes the forecasts and information about how theywere generated. Select Total>Muffins>Stuff-Mart . Notice that both views areimmediately updated to match your current selection.

For this example we used expert selection to create the forecasts. Expertselection can be thought of as an automatic pilot. It analyzes your datastatistically to determine the best forecasting technique to use and then proceeds to build the forecasts.

Expert selection begins by running a series of statistical tests on the data. Theresults of these tests are run through a rule-based logic system. At times thislogic system will directly select an appropriate forecasting method. At other times the rule-based logic will narrow down the potential forecasting methodsthat could be used without producing a clear winner. In these instances, expertselection will use an out-of-sample test to select between the competingmodels.

The Expert Analysis section of the Forecast Report documents how ForecastPro Unlimited makes its decision. In our example, the rule-based logic was



Viewing the Forecast Report 27

inconclusive, but narrowed down the choices to either an exponentialsmoothing model or a Box-Jenkins model. An out-of-sample test was thenused to choose between these two candidates.

Notice that the expert system acts in much the same way as an expertstatistical analyst. It performs a series of statistical tests, interprets the results, performs more tests if necessary and finally makes recommendations based

upon the results.

Expert selection works well for most applications. However, Forecast ProUnlimited also allows you to customize your script by adding modifiers to the Navigator. A modifier on a particular item gives Forecast Pro Unlimited instructions on how to create the forecasts for that item. We will explore theuse of modifiers in Lesson 2.

Viewing the Forecast Report

Let’s take a closer look at Forecast Report view. Turn off the Graph view byclicking its green icon ( ).

You are currently looking at a standard format Forecast Report. The ForecastReport can be customized and can include up to nine sections. They are:



28 The Basics

Expert Analysis documents the logic behind the expert selection’s decision.This section will be omitted if a user-specified forecasting model is used.

Model Details documents the specific forecasting model that was used togenerate the forecasts.

Within-Sample Statistics provides a set of standardized model statistics that

can be used to diagnose the current model as well as to compare and contrastalternative forecasting approaches.

Historic Data (not currently active) lists the history and fitted values alongwith some aggregated totals and summary statistics.

Forecast Data lists the forecasts and confidence limits along with someaggregated totals and summary statistics.

Safety stocks (not currently active) lists the safety stocks for different lead times along with the demand during lead time and reorder point.

Out-of-Sample Rolling Evaluation and Out-of-Sample Static Evaluation (notcurrently active) list the out-of-sample statistics that are generated when aholdout analysis is performed.

Outliers (not currently active) lists any detected and/or corrected outliers.

Overrides displays any overrides and/or comments that were entered.

Right click on the Forecast Report view to invoke its context menu. Select

Forecast Report Design. Notice that this dialog box allows you to select astandardized report style or create a custom one. Select Forecast Only and click OK.



Customizing the Graph View 29

Notice that this report is much simpler that the standard one. Return to theForecast Report Design dialog box and experiment with the settings. After you are comfortable with their operation edit the settings to display theStandard report style and click OK.

Customizing the Graph View

Let’s now explore some of the graphical options. Display the Graph view byclicking its green icon ( ). Turn off the Forecast Report view by clicking its

green icon ( ). Your display should now match the one shown below.

The green line on the graph represents the demand history for Total>Muffins>Stuff-Mart . The red line is the forecast and the blue lines are

the confidence limits.

Move the cursor to a point on the graph, hold the mouse very still and click the left mouse button. Notice that the display now includes a vertical red lineand a box listing the corresponding date and numeric values of the graphvariables.

Click January 2010 (the months are marked along the X-axis). Notice that thered line and box have moved to the new location.



30 The Basics

Click the right mouse button to invoke the Graph view’s context menu. Select

Data Only. Notice that selecting Data Only turned off all of the forecast-

related components (e.g., the forecasts and confidence limits). Select Data

Only a second time to turn the forecast-related components back on.

Click the right mouse button to redisplay the context menu. Notice that eachindividual graph component (History, Corrected History, Forecast,Confidence Limits, Statistical Forecasts, Fitted Values and Historic Intervals)can be toggled on or off.

Select Graph Settings. The Graph Settings dialog box allows you tocustomize your graph in numerous ways.

The Include tab allows you to toggle the forecast-related components on and

off. Select the Components tab.



32 The Basics

Click the green Override Forecasts icon ( ) or select View>Overrides from

the menu. Your display should look like the one below.

The Override view provides several different ways to override the statisticalforecasts. You can enter values in individual override cells, you can select arange of cells and use the Percent, Increment or Value buttons to write thedesired overrides to the target override row or you can use the mouse to drag points on the graph. If a time period has overrides in both Override 1 and Override 2, Override 2 will take precedence.

Highlight the Statistical values for August and September of 2009. Set thePercent box to 12% and click the Percent button. Notice that the proposed overrides are written to the target row indicated to the immediate left of theCommit button—in this example, Override 1. Notice also that the values

displayed in the Forecast row have not changed—this is because we have notyet committed to the overrides. Uncommitted overrides are displayed in red, both in the override row and on the graph.

Click the Commit button to accept the overrides. Notice that the formerlyuncommitted values in the override row change to black, the Forecast row isupdated and the graph is updated.

Notice also that the icon for Total>Muffins>Stuff-Mart on the Navigator nowappears in red and the icons for Total>Muffins and Total in yellow. A red Navigator icon signifies that an override has been made to the item. If a group



Working With a Hot List 33

icon appears in yellow it signifies that an override has been made to a variablelower down that branch of the hierarchy.

When you make an adjustment to an item in a hierarchy, Forecast ProUnlimited will automatically reconcile the entire hierarchy to reflect thechange. Thus when we made our override to Total>Muffins>Stuff-Mart thisimpacted the groups above our selection and all groups and items below our

selection. The Reference section of this manual describes how thereconciliation is performed.

Important: If you restrict your overrides to a single level of the hierarchy, the

reconciliation process is simple and intuitive. We strongly urge you to limit

your overrides to a single level if at all possible.

We have illustrated the basics of the override facility. There is a considerableamount of additional functionality, including the ability to associatecomments with override cells, the ability to add and remove override rows,the ability to name override rows, the ability to make overrides on the graph

using your mouse, the ability to enter overrides as formulas, the ability toview the history and the ability to make incremental adjustments to theStatistical Forecast row rather than overriding the values. All of thisfunctionality is described in the Reference section of this manual.

Turn off the Override view by clicking its green icon ( ).

Working With a Hot List

The Hot List is the (currently empty) box, located in the bottom left-hand corner of the screen, below the Navigator. Placing items on the Hot Listallows you to efficiently navigate and work with a subset of your items.

Right click on the Navigator to invoke the Navigator’s context menu. Select

Expand All to fully expand the Navigator. Drag Total>Muffins>Stuff-Mart from the Navigator into the Hot List area. Your display should now look likethe one below.



34 The Basics

Click on the different items in the Hot List and notice that as you do so, the Navigator selection and views are immediately updated to match your selection.

Right click in the Hot List area to call up the Hot List’s context menu. Noticethat there are a full range of options to control how Hot List items are

displayed and to remove items. Select Remove>All Items to clear the HotList.



Saving the Forecasts 35

Now select Total>Muffins>Stuff-Mart on the Navigator and right click to

bring up the Navigator’s context menu and select Add to Hot List. Noticethat there is a full range of options to add items to the Hot List. Children,

Parents and Siblings refer to one level down on the current branch, one levelup on the current branch and the same level on the current branchrespectively. In the “Analysis” menu selection, Overrides refer to items withdirect overrides and Affected refers to items with indirect overrides (i.e., items

where a direct override elsewhere in the hierarchy changed the items’forecast).

Experiment with adding and removing items from the Hot List until you arecomfortable with its operation.

Clear the Hot List, select Total>Muffins>Stuff-Mart on the Navigator, invoke

the Navigator’s context menu and select Add to Hot List > This Level. Your screen should now match the display below.

Saving the Forecasts

Forecast Pro Unlimited provides two different ways to output your forecasts.Each is described below.

The numeric output file(s) is used to save the forecasts and other informationin a concise format. If you plan to import the forecasts into a database or a



36 The Basics

planning system—this is the file you should use. You have a great deal of control over the content and format of this file and can save it in text,spreadsheet, ODBC, or XML formats.

The Forecast Report is an Excel file containing a separate forecast worksheetfor each item forecasted. Each report worksheet will contain the currentlydisplayed graph for the item and the currently displayed forecast report. Thus,

in addition to allowing you to view information on screen, the graph view and the forecast report view also are used to design the Excel-based forecastreport.

Click Project>Export. Notice that both of these output files can be saved for either all items forecasted or just the items on the current Hot List.

Arrange your display so that it includes both the Graph view and the ForecastReport view. Right click on the Hot List to display its context menu and select

Save Forecast Report. A dialog box will appear allowing you to name theExcel file. Name the file Hot List Report and save it.

Examine Hot List Report.xls in Excel and verify that it contains a worksheetfor each item on the Hot List and that the contents and format matches thecurrently displayed graph and report views.



Saving the Forecasts 37

Close the Graph window and the Forecast Report window. Open the NumericOutput Preview window by clicking the yellow Preview Numeric Output icon( ).

The Numeric Output Preview window displays the contents and format of the

currently specified numeric output file. Click Settings>Numeric Output

Design.

This dialog box allows you to design the Numeric Output. The first tab allowsyou to specify whether you wish to limit the output to time series elementsonly or include model details and summary statistics. It also allows you tospecify the file type and control the layout. The Forecast Output tab and theStatistical Output tabs allow you to specify the contents.

Explore the options on this dialog box. If you are unsure of the function of any selection click the Help button for a description.



38 The Basics

The settings you select will be used for your current project. If you click theSet as Default button on the displayed tab, the current settings will also besaved as the default to use for all new projects. Exit the dialog box.

Click the purple Save Numeric Output icon ( ). A dialog box will appear

allowing you to name the file. Name the file Test Numeric Output and save it.Examine the file and verify that its contents and format matches those

specified in the Numeric Output Design dialog box.

Saving the Project

Our final step will be to save our current forecasting session as a project .Saving a forecast project allows you to return later and pick up where you leftoff or to share the session with others.

Select File>Save As and save the project using the name test project .

Be aware that a forecast project saves the data, forecasts and overrides that arecurrently in memory. If you update your input data files after saving a project,when you reopen the project you’ll have the option of either opening the project as it existed when you saved it or reading in the updated data and revising the forecasts.

Exit the program.

This concludes Lesson 1.



39

Lesson 2

Using Forecast Modif iers

In the very first lesson you prepared forecasts using expert selection— Forecast Pro Unlimited’s default model selection procedure. In this lesson youwill use forecast modifiers to dictate the forecasting models to be used for specific items.

Specifying Forecast Modifiers for Individual Items

Start up the program and click the red Script icon ( ) to call up the Define

Script dialog box. Double click on ACME Hardware to place it on the first

line of the script. Click OK to accept the script. Click the red Read Data icon( ) to read in the data and then the red Forecast icon ( ) to generate the

forecasts.

Fully expand the Navigator using the Expand All option on its context menuand open the Graph view by clicking its green icon ( ). Select

Total>USA>PG1-204-2. Your display should now match the one shown below.



40 Using Forecast Modi fiers

If you do not instruct Forecast Pro Unlimited to use a specific forecastingtechnique it will use expert selection to select the appropriate technique for each item forecasted. If you examine Total>USA>PG1-204-2’s forecast

report you’ll notice that it includes an Expert Analysis section and that expertselection chose to forecast this item using a nontrended, seasonal exponentialsmoothing model.

Let’s say that we wish to dictate that a Winters exponential smoothing model be used for this item. Winters is a form of exponential smoothing which

includes both a trend and seasonal component.

To specify the model, right click on Total>USA>PG1-204-2 to bring up the

Navigator’s context menu and then select Model>Exponential

Smoothing>Winters.



Specifying Forecast Modifiers for Multiple Items 41

Notice that \WINTERS now appears next to Total>USA>PG1-204-2 on the Navigator. \WINTERS is a forecast modifier. In this instance it is used toindicate that you have opted to use a Winters exponential smoothing model

for this item rather than expert selection.

Examine the forecast report to verify that the Winters model was used. You’llnotice that the Expert Analysis section is no longer present and that the ModelDetails section indicates that a “User Defined \WINTERS” model was built.

Select Total>USA>PG1-204-4 on the Navigator. Right click to call up the

context menu and select Model>Very Simple Models>Same as Last

Year>Plus Percent. Specify the percentage as “10” and click OK.

Notice that \SALYP=10 now appears next to Total>USA>PG1-204-4 on the Navigator. This model will set the forecasts to equal the same values as the preceding year plus 10%.

Specifying Forecast Modifiers for Multiple Items

Select the group Total>Canada on the Navigator. Right click to call up the

context menu and select Model>Box-Jenkins>Auto. Notice that \BJ nowappears to the right of the group Total>Canada.



42 Using Forecast Modi fiers

Let’s say that we wish to dictate that a Box-Jenkins model be used for every

item in the group Canada. To accomplish this, select Apply Modifier(s)

To>Children on the Navigator’s context menu. Notice that the \BJ modifier now appears on all of Total>Canada’s “children” (i.e., all items one leveldown in the group Total>Canada).

Specifying Custom ForecastModifiers

In addition to the commonly used forecasting models found on the Modelcontext menu, Forecast Pro Unlimited supports a broad range of custommodeling modifiers. These modifiers accommodate a wide range of modelingoptions including confidence limits percentiles, safety stock lead times and power transformations.

Custom modifiers are entered using the Custom Modifier option on the Navigator’s context menu.

Consult the Using Forecast Modifiers section to see a full list of the availablemodifiers. The modifiers are also documented in the help system.

Experiment with the various modeling options until you are comfortable withtheir operation. When you are finished, exit the program.




43

Lesson 3

Collaborating With Colleagues

Corporate forecasting often requires collaboration among colleagues. Perhapsyour manager wishes to review your forecasts and make a few changes.Perhaps you want to run your forecasts past the sales force who may knowabout pending orders. Perhaps your business has distinctly different marketsthat require specific individuals to work on specific markets. Whatever thereason, it is often useful to be able to work with others to establish the finalforecasts.

In this lesson we will explore how Forecast Pro Unlimited supportscollaborative forecasting.

Sharing a Stand-alone Project

In this lesson we will open a forecast project that was created by someoneelse, review their work and make some overrides.

Providing a colleague access to your forecast project is a simple way to shareyour work. As long as your colleague has Forecast Pro Unlimited (Forecaster license or Collaborator license), he or she can open the project and continuethe session.



44 Collaborating With Colleagues

When you save a forecast project, Forecast Pro Unlimited creates three files.If the project name is Sample the three files are:

Sample.FPProj is the project settings file. The project settings file is an XMLfile containing the project settings and script definition.

Sample.mdb is the project tables file. The project tables file is a Microsoft

Access database file containing separate tables for the in-placeoverrides/comments, forecast modifiers and Hot List.

Sample.fcb is the project snapshot file. The project snapshot file is a binaryfile that saves the forecast components that are currently in memory (e.g., theforecasts, summary statistics, etc.) so that the session can be restored withouthaving to read in the data, recalculate the forecasts, etc. This allows you toreturn to your session where you left off and to share the session with others.

When sharing projects with others, you need to provide all three files.

To illustrate how to share a project we’ll open up a project that is included with Forecast Pro Unlimited. Start Forecast Pro Unlimited, select File>Open

and open the project, Tutorial 3. The data in this project represent sales of different types of cameras and film.

Notice that we did not define a script, read the data or click the forecast button. When we opened the forecast project, Forecast Pro Unlimited restored



Sharing a Stand-alone Project 45

the script, forecasts, overrides and Hot List that were present when theforecast project was saved. The person restoring the forecast project does noteven need to have access to the files containing the historical data that wereused to generate the forecasts—the only thing he or she needs is the forecast project files.

Notice that the person who created the forecasts entered some overrides in

June, July and August. For convenience, let’s refer to the person who created this file as “John”. Notice also that the background color for these cells isyellow. This indicates that the cells contain comments. Click on the Juneoverride cell to display the comment.

The comment informs us that John increased the forecast for the summer months by 20% in anticipation of a planned promotion. Let’s say that we feelthat a 20% lift is overly optimistic for August (the final month of the promotion) and we want to lower it to 15%. Set the target row box (located tothe immediate left of the Commit button) which currently reads John toOverride 2. Highlight the cell for August on the Statistical forecast row. Set

the percent box to 15 and click the Percent button. Let’s now add a comment.Highlight the cell for the override you just made, click the comment field and type, “John, I lowered the promotional lift for summer madness to 15%” and click the Commit button to accept the overrides and comment.

Now let’s edit the row label Override 2. Click on the row label Override 2 and

right mouse click to call up the context menu. Select Edit Row Label. Enter Management and click OK. Notice that the row label has changed.

At this point we have several options on how we might save our work. For instance, we could:

Use File>Save as to save a new forecast project to send back to John.

Click the purple Save Numeric Output icon ( ) to save an output file

to submit to production, management or some other destination.

Click the purple Save Forecast Report icon ( ) to save formatted

reports to Excel.

Exit Forecast Pro Unlimited without saving changes to the Tutorial 3 project.



47

Lesson 4

Updating a Forecast Project

For most organizations updating the forecast is a routine operation that occursevery planning period. For instance, if the forecasts are prepared usingmonthly data, then each month the data are updated to include the latestobservation and the forecasts are regenerated.

Forecast projects enable you to quickly update last period’s forecast and giveyou the option of loading or not loading your previously specified forecastingmodels, Hot List and forecast overrides. In this lesson we will explore howthis is accomplished.

Defining the Forecasting Process

An important key to implementing a successful forecasting process usingForecast Pro Unlimited is to define the process upfront and make sure thateveryone involved understands their role in establishing the final numbers.

Some of the more important questions to answer before you start include:

What output is required and what format should it be in?



48 Updating a Forecast Project

What naming conventions should be used for data files, output filesand forecast projects, where should these files be kept and how willthey be backed up?

Who is authorized to enter forecast overrides and at what level(s) arethey entered?

What are the steps required to update our forecasts each planning period?

Depending on the complexity of your forecasting process and the number of individuals involved, it may be useful to document the decisions made and thesteps required to update the forecasts each planning period.

Step One: Updating Your DataFiles

Consider the following situation. It is early June 2009 and we have historicdemand data that starts in 1/2004 and ends in 5/2009. Thus, our first forecast period will be June 2009. We use Forecast Pro Unlimited to create our forecasts and then we save a forecast project named Tutorial 3-June 2009.

Now imagine that a month passes by. We now have our sales figures for Juneand we update our historic data files.

This is accomplished externally to Forecast Pro Unlimited. It may entailrunning a data extraction routine to generate the new files, updating your spreadsheets by hand, or some other process to update the historical data files.

Step Two: Opening Your ForecastProject

Start Forecast Pro Unlimited, select File>Open and select the project, Lesson

3-June 2009. The dialog box below appears.



Step Two: Opening Your Forecast Project 49

Forecast Pro Unlimited has noticed that the forecast project Lesson 3-June

2009 and the data file 123 Bakery.xls are out-of-sync. It is giving you twooptions.

If you select “Do not recalculate the forecast”, Forecast Pro Unlimited willopen the project and restore the forecasts session as it existed when it wassaved last month. It will not read the new data. It will not revise the forecasts.

Select “Do not recalculate the forecast” and click OK to verify this. Noticethat the historic data ends in May of 2009 and the first forecast period is June2009.

Now select File>Open and select the project, Lesson 3-June 2009 again. Thedata file and project are still out of sync so the dialog box reappears.

If we now select “Recalculate the forecast and apply”, Forecast Pro Unlimited will read in the revised data, generate new forecasts and (if you request them)restore the project settings, Hot List, forecast modifiers and overrides.

Select “Recalculate the forecast and apply” and click OK. Expand the Navigator and select Total>Muffins>Stuff Mart>BRA-12-11.

Notice that the historic data now ends in June 2009 and the first forecast period is July 2009. Notice also that Forecast Pro Unlimited has restored the

Hot List, forecast modifiers and overrides and that the overrides correctlyreflect the new dates.

At this point we have generated a new forecast. Let’s save this project using adifferent name so that we can retain both our Tutorial 3-June 2009 project and

our new July forecasts. Select File>Save As and name the project Tutorial 3-

July 2009.



50 Updating a Forecast Project

Step Three: Saving Your Work

If this was a real forecasting session we would now review the new forecasts,add any needed overrides, adjust the forecasting models as appropriate, sharethe forecast project with others, etc. until we establish the final July forecast.

Once we’ve established the final forecast we would save our final project filefor the July forecast and save any desired output files.

Exit Forecast Pro Unlimited using File>Exit.




51

Lesson 5

Building Event Models

Event adjustment models extend exponential smoothing by allowing you toadjust for events like sales promotions, strikes or simply for unexplained outliers. You can adjust for events of several different types. These could be promotions of different types or sizes, or different calendar effects like Easter and Independence Day weeks. Forecast Pro Unlimited knows theseoccurrences simply as events of types 1, 2, etc.

Event adjustment models work almost the same as seasonal index models. In aseasonal index model, each month gets its own index, which is updated eachtime that month recurs. In an event adjustment model, each event type gets its own index, which is updated each time an event of that particular type recurs.The difference is that while January recurs every 12 months, an event of type1 usually recurs irregularly.

Since Forecast Pro Unlimited knows that January occurs every 12 months,you don't have to provide that information. To implement event adjustment,however, you must tell the program when events of each type occur. To dothis, you must construct an event variable which classifies each period byevent type (0=no event, 1=event of type 1, 2=event of type 2, etc.). Theformat is the same as that for any other historic data record, except that itsvariable name must begin or end with an underscore ( _ ) and its data entriesmust all be integers. The event variable must be defined for each period in the



52 Building Event Models

historic record. If you want to forecast the effects of future known events, youmust include these future periods as well.

For more statistical details, please consult the statistical reference for exponential smoothing in the Forecast Pro Statistical Reference Manual.

In this lesson you will use an event model to capture the relationship between

sales of mouthwash and promotions. You will also use an event model tocapture seasonal patterns in weekly data.

Modeling Promotions

Click the red Script icon ( ) and create a script listing Minty Fresh

Mouthwash on the first line. Click OK to accept the script, then click the red Read Data icon ( ) to read in the data and, finally, click the green View

Graph icon ( ) and use the Navigator to display a graph of MintyFresh24.

MintyFresh24 represents monthly sales of a nationally advertised brand of mouthwash in a certain size package. The prominent peaks on the graph are

not due to seasonal patterns ⎯ they are the result of price promotions. SelectMintyFresh32 on the Navigator. This series represents sales of the same product in a different size package. “_PriceSpecials” is an event variable—wewill discuss this variable shortly.



Modeling Promotions 53

Let's begin by modeling MintyFresh24 and MintyFresh32 using expertselection. Click the red Forecast icon ( ). When the calculations are

complete, view the forecast report for MintyFresh24. Notice that a nontrended additive seasonality exponential smoothing is selected and that the adjusted R-square is 0.50. Exponential smoothing recognized the effects of the promotions as a form of irregular seasonality. The forecasts from such amodel tend to repeat the promotional pattern of the last year of historic data.

If, as is usually the case, your future promotions are patterned differently, theforecasts may be very bad.

Include the fitted values on the graph. (This is accomplished by toggling onFitted on the graph’s context menu). Notice that the model predicted several peaks that did not occur as well as missing several that did. The model istrying to capture the peaks as part of the seasonal pattern. However, the promotions that caused these peaks did not always fall in the same months and thus cannot be modeled using a seasonal model.

Select _PriceSpecials on the Navigator.

_PriceSpecials is an event variable. _PriceSpecials takes the value of zerowhen MintyFresh24 and MintyFresh32 are not promoted and the value of onewhen they are. Notice that the series includes the promotional schedule for theforecast period. When you use _PriceSpecials to build an event model anadjustment will be included for the promoted months (any variable whose




name begins or ends with an underbar is recognized as a “helper” variable and will not be forecasted).

_PriceSpecials reflects the simplest type of event variable. Each month iscoded either as promoted (1) or not promoted (0). At times you may havemore than one type of event. For example, imagine that you had both price promotions and rebate promotions. In this case you would create an event file

coding months as not promoted (0), price promoted (1) or rebate promoted (2). When you build the event model it will include two different adjustments,one for price promoted months (type 1 months) and one for rebate promoted months (type 2 months).

Select MintyFresh24 on the Navigator. Right click to call up the context menu

and select Events. Selecting Events will display the available helper variables

and allow you to select the one you wish to use. In this example there is onlyone helper variable, _PriceSpecials. Select _PriceSpecials to build the eventmodel.

Examine the forecast report for MintyFresh24. Notice that an exponentialsmoothing model was used and that the model includes an event smoothingweight. In addition, notice that the adjusted R-square is now 0.79.

The graph the results reveals a much better fit to the data and forecast peaksthat correspond to the months indicated as promoted rather than the“seasonal” pattern.



Modeling Weekly Data 55

Select Apply Modifier(s) to on the Navigator’s context menu and select All

Items. This will specify that the event model should be built for every item onthe Navigator. In this example, this is appropriate since the same promotionalschedule was used for both MintyFresh24 and MintyFresh32.

If the two products had been promoted differently we would have created twoseparate event variables.

Business Forecast Systems has conducted extensive out of sample testing onevent adjustment models. This research demonstrates that event adjustmentmodels nearly always yield a substantial real world benefit in forecastaccuracy.

Modeling Weekly Data

Select File>New. The program will ask you if you wish to save changes to the

current project—click No. Click Settings>Options, set the forecast horizon to

52 and click OK. In this exercise we will model weekly beer sales for a brand of beer in various types of packages.

Create a script containing the two lines

Beer Beer Events

In the last example, our event schedules were included directly in our datafile. In this example, we’ve elected to put the event variables in a separate filenamed Beer Events. Maintaining your event schedules and/or weightingvariables (discussed in Lesson 11) in a separate file from the historic data isoften convenient. For instance, if your IT department generates the historic




data files for you each forecast period, keeping the event schedules in aseparate file allows you to edit and maintain them without the assistance of the IT department.

Read in the data, create the forecasts, expand the Navigator and graph C-6. C-6 represents sales of beer in cans packaged in 6-packs.

In order to get a better view of the data, turn off the Forecast Report window.

Right click the Graph view to invoke its context menu and select Graph

Settings. Open the Components tab, remove the checkmark for “All”, set“Periods” to 156 and click OK. Note that the display is now scrollable.Periods allows you to select how many data points will be shown on-screenwhen the All option is not selected.

Add the fitted values to the graph using the graph’s context menu.



Modeling Weekly Data 57

One noticeable feature of this data set is that sales for the weeks containingthe three US summer holidays (Memorial Day, 4th of July, Labor Day) arehigher than other weeks. If you look closely at the data you’ll notice that these

holidays sometimes change weeks, and when this occurs, the fit can be poor.For instance the 4th of July fell in week 27 before 2006, but fell in week 28 in2006 and thereafter. The fitted value for week 27 in 2006 exhibits a strong peak and “misses” the actual holiday peak that fell in week 28.

Forecast Pro Unlimited used a standard Winters model to forecast C-6. This

model uses 52 seasonal indexes for the data and does not accommodate theholidays changing weeks from year to year.

Graph the event variable _Holidays. _Holidays takes on the value of “1” for the week containing Memorial Day, the value of “2” for the week containing

the 4th of July, the value of “3” for the week containing Labor Day and “0”for all other weeks. Including an event index for each holiday allows us tomodel the holidays that move around the calendar. (If we were to model thisdata as a monthly series, the holidays would not change periods and an eventmodel would not be necessary.)

Build the model by clicking C-6 on the Navigator and using the context menuto add the event variable _Holidays.




The model now includes 52 seasonal indexes to capture seasonality and threeevent indexes to capture the summer holidays. Examine the fit to the holidayweeks and notice that the event model is not “missing” the holidays when they

move around the calendar.

Although we would like the seasonal indexes to capture a smooth seasonal pattern in this case they do not. This is because of the low level of temporalaggregation and other problems peculiar to weekly data:

The year usually begins and ends with a partial week.

Holidays such as Easter and Independence Day do not occur in thesame week each year.

Some years may contain 53 weeks, depending upon how a week is

defined. Different corporations address this differently.

Generally speaking, monthly models will more accurately capture seasonality. Nevertheless, many corporations must deal with weekly seasonal data because production and inventory control require it. Event adjustment models can helpyou with some of the problems that you will face.

Event models allow you to control how many indexes to include in the model.To illustrate, let's model C-6 using the _P&H event variable.



Adjustment for Out liers 59

_P&H maps each week into one of 13 periods (i.e. weeks 1-4 are coded 1,weeks 5-8 are coded 2, etc.). The summer holiday weeks (Memorial Day, 4

th

of July and Labor Day) are coded as 14, 15 and 16 respectively. You maywant to examine the data in the Beer Events.xls spreadsheet to see preciselyhow this is accomplished.

Turn back on the Forecast Report view and then use the Navigator’s context

menu to specify that _P&H be used as the event schedule for C-6.

The resulting model uses 16 event indexes—13 to capture the seasonality and three for the summer holidays.

Notice that the forecasts exhibit much smoother seasonality than our previousmodel. That is because there are now more samples of each index to beestimated.

Adjustment for Outliers

Historic data sets often exhibit the effects of one-time events that causeoutliers. The event that causes the outlier may be known or unknown.Although exponential smoothing is a remarkably robust procedure, theseoutliers may decrease the quality of the forecasts and (especially) theconfidence limits.




You can eliminate the effect of an outlier by coding it as a special event thatoccurs only once. If you have several outliers, each must be coded as adistinct event type. Forecast Pro Unlimited will “explain” each outlier as theresult of its associated event.

The impact of outliers on the forecasts and the confidence limits will begreatly reduced. Beware however: if outliers continue to occur in the forecast

period, then the confidence limits are likely to be unrealistically narrow.

More Adjustments for Promotions

The event adjustment model provides a very flexible framework to treat promotional effects of many kinds. This section briefly notes some of them.Its purpose is to indicate some directions you may want to take with your own business data.

The examples that are given require coding additional event types in your

event variable. Each new event type provides additional ability for ForecastPro Unlimited to explain your historic data by making the event variable morecomplex. Keep in mind that if your event description is overly complex, theout-of-sample performance of your model may deteriorate. You must strikethe right trade-off between goodness-of-fit to your historic data and modelcomplexity. To do this will require experimentation and monitoring of actualmodel performance.

Example 1. A promotion in (say) September may have effects in August and October as well. Buyers may delay purchases in August and they may beoverstocked in October. You can code the pre- and post-promotional effectsas event types of their own. These events will, of course, be associated withdecreases in sales.

Example 2. Sometimes one SKU of a brand or product line is promoted butclosely related SKU's are not. The result may be that the promoted SKUcannibalizes the sales of the other SKUs. You can treat this effect by codingcannibalization events for these SKUs. But be cautious. Overuse of thistechnique will result in an overly complex model and possible deterioration of forecast performance.




61

Lesson 6

Building Multiple-Level Models

In this lesson, you will learn how to define group variables and preparemultiple-level forecasts.

Bottom-up and Top-downForecasting

Start the program, click the red Script icon and select Presto Camera

Company. Accept the script, read in the data, forecast the data, open the graphview and fully expand the Navigator. Your screen should now look like theone shown below.



62 Building Multiple-Level Models

Notice that the Navigator tree structure matches the hierarchy shown below.

Total

Camera Film

vCamera iCamera vFormat_Film iCamera_Film Etc.

The tree contains three group variables (Total, Camera and Film). The treealso contains six end items (vCamera, iCamera, vFormat_Film,iCamera_Film, ASA100_BW and ASA100_Color). The demand histories for group variables are not in the database. They are created by Forecast ProUnlimited by aggregating the appropriate end items. Thus, the history for thegroup Camera is defined as the sum of vCamera + iCamera. The history for

the group Film is defined as the sum of vFormat_Film + iCamera_Film +ASA100_BW + ASA100_Color. The history for the group “Total” is defined as the sum of the groups Camera + Film.

When you forecast a multiple-level hierarchy Forecast Pro Unlimited willautomatically reconcile the forecasts. By default a bottom-up reconciliationwill be performed (this is the case in our current example). That means that

the group-level forecasts will be generated by aggregating their componentforecasts. For example, the forecast for the group Camera would be made bysumming the forecasts of vCamera and iCamera.



Bottom-up and Top-down Forecasting 63

An alternative is to use top-down reconciliation. Select Camera on the Navigator and right click to display the Navigator’s context menu. Select Topdown. Notice that the \TOPDOWN modifier now appears next to Camera.

The \TOPDOWN group modifier instructs the program to adjust lower-levelforecasts to sum up to the indicated level. Let’s illustrate how this works byconsidering the various steps involved in forecasting our current example.

Step One: Prepare forecasts at all levels (i.e., prepare forecasts for “Total”,Camera, Film, vCamera, iCamera, vFormat_Film, iCamera_Film,ASA100_BW and ASA100_Color based on their past history).

Step Two: Since the group Camera is being forecasted top down, the forecastsof vCamera and iCamera are adjusted proportionally so that they sum to theCamera forecast.

Step Three: Since the group Film is being forecasted bottom up (the default)the Film forecast is replaced with the sum of the forecasts for vFormat_Film,iCamera_Film, ASA100_BW and ASA100_Color.

Step Four: Since the group “Total” is being forecasted bottom up (the default)the “Total” forecast is replaced with the sum of the forecasts for Film +Camera.




There is one other group modifier. \INDEXES instructs the program toestimate the seasonal indexes at the group level and use them for all membersin the group. This is particularly useful when some members of the grouphave short histories or small demand and seasonal factors are thus difficult tocalculate.

Defining Groups Using Attr ibuteFields

The spreadsheet Presto Camera Company.xls is shown above.

In this example, we defined the groups Total, Camera and Film directly in thespreadsheet using attribute fields (columns A and B).

If we had included additional attribute fields we would have created a larger hierarchy. Consult the Setting Up Your Data section of this manual for complete details.



Combining Multiple Level and Event Adjustment Models 65

Combining Multiple Level andEvent Adjustment Models

In many cases, the SKU's or product lines in a multiple-level hierarchy aresubject to promotional effects. You can use the event model methodology

described in Lesson 5 to account for such effects.

Example 1. A product line, consisting of aggregated SKU's is promoted as aunit. You believe that all the SKU's in the product line will be affected similarly. You can deal with this by using the following approach.

LINE \EVENT=_PROMO \TOPDOWNSKU1

SKU2SKU3

This causes Forecast Pro Unlimited to execute the following procedure.

Forecast the group LINE, taking promotions into account.

Forecast the SKU's without taking promotions into account.

Adjust the SKU forecasts so that they sum to the LINE forecasts. This propagates the group level promotional effects to the SKU's.

Example 2. Now suppose that only one SKU of the group is promoted. Thefollowing approach can account for this promotion.

LINE

SKU1 \EVENT=_PROMOSKU2SKU3

This causes Forecast Pro Unlimited to forecast the SKU's individually. It thensums these forecasts to obtain forecasts for the group LINE.

What if you were to add the keyword \TOPDOWN to the group LINE in thisexample? The effect would be that the promotion of SKU1 affects it but doesnot affect overall sales at the group level. This might be the case if the




promotion affected sales only by cannibalizing other SKU's in the group. Thisis a dubious effect. It is usually better to prepare forecasts bottom-up whenthere are distinct models for the lower-level units.

Short-Lived Products

A manufacturer often maintains a product line for a long period of time butfrequently changes the SKU's that make up the line. A laser printer manufacturer, for instance, often introduces new models and retires old ones.The result may be that the overall product line can be accurately forecasted, but the individual item histories are too short to support seasonal models. Inthese instances the top-down approach is particularly useful.

For instance assume that SKU1 and SKU2 have been phased out and replaced by SKU3 and SKU4. We also assume that at the end of the historic data, onlySKU3 and SKU4 are alive. However, the histories for SKU3 and SKU4 aretoo short to generate seasonal forecasts. The manufacturer is interested in

forecasting the group LINE, SKU3 and SKU4. The approach below takes careof the problem.

LINE \INDEXESSKU1SKU2SKU3SKU4

This causes Forecast Pro Unlimited to follow the following procedure.

Forecast LINE, obtaining seasonal indexes.

Use the LINE seasonal indexes to deseasonalize the SKU's.

Forecast the resulting nonseasonal SKU-level data. These nonseasonalmodels require very little data.

Use the LINE seasonal indexes to reseasonalize the SKU-levelforecasts.



Short-Lived Products 67

By default, the LINE forecasts are then replaced by the summed SKU-levelforecasts. If you do not want this to happen, you can add the keyword \TOPDOWN to the group LINE.

The script presented in this example can also be used when the SKU-levelhistories are long lived. The result is that seasonality is accounted for at theLINE level. This is desirable when the SKU's are likely to have similar

seasonal patterns, but the data are too irregular for accurate estimation of seasonal indexes at the SKU level.




69

Lesson 7

Using Weighting Transformations

Forecast Pro Unlimited includes a weighting transformation which will divideyour historical data by user-defined weights, forecast the resultant(deweighted) series and then multiply (reweight) the forecasts. This procedurecan be useful in a variety of situations including adjusting for the number of working days in a month, defining a seasonal pattern and supplying a growthcurve for a new product.

In this section we will use the weights procedure to adjust for the number of weeks in each month. The section will conclude with a discussion of how toapply weights in other situations.

Adjusting for 4 vs. 5 WeekPeriods

It is not uncommon for companies to divide the year into twelve periods eachconsisting of either four or five complete weeks. This is often referred to as a4-4-5 calendar. In situations where the forecasts ultimately need to be brokendown by week, this method may be preferable to using actual calendar monthswhich include partial weeks.



70 Using Weighting Transformations

If the number of weeks in each period is consistent from year to year (i.e., period 1 always has 4 weeks, period 2 always has 4 weeks, period 3 alwayshas 5 weeks, etc.) then the variation due to the number of weeks in each period will be captured as part of the seasonality and no special action isrequired. In cases where the number of weeks in each period varies from year to year (i.e., some years period 1 has 4 weeks other years it has 5) a weightingtransformation can be used to capture the variation due to how many weeks

are in each period.

The procedure is very straightforward. You construct a weighting variablethat classifies each period of the historical data and forecast period ascontaining either 4 or 5 weeks. Forecast Pro Unlimited divides the historicaldata by the weights. (This converts sales per period into sales per week per period.) This deweighted series is forecasted. The forecasts are thenreweighted by multiplying by the corresponding weights.

The data for this exercise are contained in two Excel files. 4-4-5 Data.xls

contains five time series that were collected using twelve 4 or 5 week periods

per year. The number of weeks per period is not consistent from year to year.4-4-5 Helper.xls contains the helper variable “_Weeks”.

Helper variables are either weighting variables or event variables. They areidentified by the use of a preceding or trailing underbar. They can either be placed in the same data file as the series they are helping or placed in their own separate file. In this example the helper variable _Weeks contains thenumber of weeks in each period. It covers the historical span as well as theforecast period.

Select Settings>Options and make sure that the data mode is set to Xls and

that the forecast horizon is set to 12. Create a script containing the two lines

4-4-5 Data4-4-5 Helper

Read in the data and click the red Forecast icon to build the models. When thecalculations are complete, make sure that the Graph and Forecast Reportviews are open and view the results for Product 1 (shown below). Notice thata seasonal model is selected and that the adjusted R-square is 0.50.



Adjust ing for 4 vs. 5 Week Periods 71

Right click on Product 1 to call up the Navigator’s context menu. SelectWeights and then select _Weeks.

The second model’s superior fit and narrower confidence limits are easilydiscernible from a visual inspection of the graph. Notice that a seasonal modelis selected and that the adjusted R-square is now 0.82.

Right click on Product 1 to call up the Navigator’s context menu. Select

Apply Modifier(s) To>All Items. Since the other four products all use the



72 Using Weighting Transformations

same calendar, applying the weighting transformation improves the model inall cases.

Other Uses for Weight ingTransformations

The weighting transformation is useful in a wide variety of situations. Thissection briefly notes some of them. Its purpose is to indicate some directionsyou may want to take with your own business data.

Trading day corrections. Many businesses are sensitive to the number of working days per period. Consider a service provider who is closed on theweekends. The number of working days in January (and all other months) willvary from year to year depending on how many weekend days happen to fallin any given month. If the number of working days has an impact on sales,then it needs to be accounted for in the model. A simple solution would be touse a weighting transformation where the weights consist of the number of

working days per month.

User defined seasonality. At times you may wish to supply your own estimateof the seasonal pattern rather than trying to extract it directly from the data.This might be desirable if the data were short or very noisy. The weightingvariable would consist of seasonal multipliers for the series. Since theseasonality is being handled by the weighting variable you may want torestrict the model selection to nonseasonal models (perhaps using the custommodel \EXSM=*N).

New product forecasting. The weighting transformation can be used to force a

new product’s forecast to mimic a certain shape. The weighting variablewould consist of the desired shape (e.g., the history of a similar product).Since the shape of the forecast is being handled by the weighting variable youmay want to restrict the model selection to nontrended, nonseasonal models(perhaps by forcing a Simple exponential smoothing model).




73

Lesson 8

Detecting and Correcting Outl iers

An outlier is a data point that falls outside of the expected range of the data(i.e., it is an unusually large or small data point). If you are forecasting a timeseries that contains an outlier there is a danger that the outlier could have asignificant impact on the forecast.

One solution to this problem is to screen the historical data for outliers and replace them with more typical values prior to generating the forecasts. This process is referred to as outlier detection and correction.

Correcting for a severe outlier (or building an event model for the time seriesif the cause of the outlier is known) will often improve the forecast. However if the outlier is not truly severe, correcting for it may do more harm than good.

When you correct an outlier, you are rewriting the history to be smoother thanit actually was and this will change the forecasts and narrow the confidencelimits. This will result in poor forecasts and unrealistic confidence limits whenthe correction was not necessary.

It is the authors’ opinion that outlier correction should be performed sparinglyand that detected outliers should be individually reviewed by the forecaster todetermine whether a correction is appropriate.



74 Detecting and Correcting Outliers

Forecast Pro Unlimited incorporates an automated algorithm to detect and (optionally) correct outliers. In this lesson we will explore its operation.

How Outlier Detection andCorrection Works

Start Forecast Pro Unlimited, select Settings>Options and display theOutliers tab.

There are three modes for outlier detection and correction.

None turns outlier detection and correction off. This is the default and manyForecast Pro Unlimited users do not use (nor need to use) outlier detection.

Detection only will detect outliers and display the suggested corrected values,however, the forecasts will be generated using the uncorrected history.

Detection and correction will detect outliers and will automatically use thecorrected values when generating forecasts.

The detection and correction algorithm works as follows:

1. The specified forecasting model is fit to the time series, the residuals (fitted errors) are generated and their standard deviation is calculated.

2. If the size of the largest error exceeds the outlier threshold, the point isflagged as an outlier and the historic value for the period is replaced with thefitted value.



Reviewing the Outliers 75

3. The procedure is then repeated using the corrected history until either nooutliers are detected or the specified maximum number of iterations isreached.

In a multiple-level problem the detection is only performed on the end items(i.e., the nongroup level). If the correction option has been selected, after allend items are corrected, the group level totals are reaggregated to reflect the

corrected values.

You can adjust the Sensitivity setting to make the outlier threshold more or less sensitive. The proper setting will depend on the stability of your data set.

Set the detection and correction mode to Detection only, make sure the other settings match the selections shown above and click OK.

Reviewing the Outliers

In this exercise we will perform the following steps:

1. Use the Detection only mode to identify the outliers.

2. Use the Outlier Report view and the Graph view to review the detected outliers and determine whether a correction is warranted.

3. Use the \OUTLIER=CORRECT modifier to correct the outliers that we feelshould be corrected.

Make sure that the data mode is set to XLS and that the forecast horizon is setto 12. Create a script containing the single line

123 Bakery

Read in the data and click the red Forecast icon to build the models. When thecalculations are complete, click the yellow View Outlier Report icon ( ) to

open the outlier report view and turn off any other open views.



76 Detecting and Correcting Outliers

Notice that a total of 6 outliers have been detected and their current Status is Detected .

Correcting the Outliers

Open the Graph view and double click on the first item listed on the overridereport. Your display should now match the one shown below.



Correcting the Outliers 77

Notice that when you double clicked an item on the outlier report, the Navigator jumps directly to that item. Thus, you can use the outlier report asthe equivalent of a Hot List to navigate through the listed items. All of theglobal report views (yellow icon views) support this kind of navigation.

Notice that the graph is displaying both the actual history for December 2007and the suggested correction. The data point is rather unusual and you can see

that the forecast has a small spike in December which is being driven by theoutlier.

Right click the item on the Navigator (not the outlier report) to bring up the

Navigator’s context menu. Select Outliers>Correct. Notice that the status inthe outlier report has changed to Corrected and that correcting for this outlier has changed the shape of the forecasts.

Double click on the next item on the outlier report. Here too, we have a fairlysignificant outlier and will want to correct for it. Examine the remaining itemson the Hot List and experiment with the Outlier options on the Navigator’s

context menu. When you are comfortable with their operation, exit the program.




79

Lesson 9

Out-of-sample Testing

A good deal of the empirical knowledge about forecasting has come fromcomparisons of different methodologies. The M-Competition (Makridakis etal. [1982]) and M-3 Competition (Makridakis and Hibon [2000]) are thelargest and most famous of these comparisons. Forecast Pro participated in theM-3 competition and outperformed all other software entrants and 16 out of 17 academic teams.

The rather simple comparison methodology for the original M-Competitionwas as follows.

The researchers assembled a collection of 1001 time series of yearly, quarterlyand monthly data. The data were obtained from microeconomic, industry-

level, macroeconomic and demographic sources. Twenty forecasting methodswere tested for the entire sample of 1001 time series, and three on a subset of only 111 time series.

A sample of time points (6 for annual series, 8 for quarterly, 18 for monthly)was held out from the end of each time series. Each forecast model was fitted to the remaining data, and used to forecast the values of the holdout sample.The forecasts were then compared to the withheld data, and errors computed for each horizon, each time series and each forecast method. The errors werethen summarized and analyzed in a variety of ways.



80 Out-of-sample Testing

The most significant weakness in this methodology is that it uses only oneforecast base for each time series, the last point in the fitting sample. Oneobtains only a “snapshot” of performance from one point in time. A forecast base just before or after a dramatic event in the data may completely changethe results. Furthermore, you obtain only one forecast error for each horizontime from 1 to the end of the fit set. This procedure is referred to as a static evaluation.

Forecast Pro Unlimited implements both a static and a rolling base evaluation.The rolling base procedure begins in the same way. However, after theforecasts have been made, the model is rolled forward by one period.Forecasts are then made from the new base to the end of the withheld data.This process is repeated until the withheld data sample is exhausted. If 6 data points have been withheld, then you obtain 6 1-step forecasts, 5 2-stepforecasts, 4 3-step forecasts, etc.

The model coefficients are not reestimated as each additional data point isassimilated. The forecast model is based entirely upon the original fit set.

Setting Up the Script for Evaluation

Forecast Pro Unlimited is shipped with data files containing the 111 seriesfrom the original M-competition. Annual contains 20 annual series, Quarterlycontains 23 quarterly series, and Monthly contains 68 monthly series.

In this lesson we will compare the out-of-sample performance of two differentforecasting techniques using the monthly data set. The main idea of this

facility in Forecast Pro Unlimited is to give you the opportunity to test optionsfor time series from your own company. This lets you tune Forecast ProUnlimited to your data.

Click the red Script icon and on the first line enter:

M-data Monthly

Accept the script and then click the red Read Data icon to read in the data. Setthe holdout sample on the dialog bar to 6. Click the red Forecast icon to createthe forecasts.



Examining the Analytic Output 81

Examining the Analytic Output

After processing is complete, select the group Monthly on the Navigator and view the Out-of-Sample Rolling Evaluation for Items in Group section of theForecast Report view.

The tables display the following information. Means are always taken over allthe variables forecasted for a particular script line.

Hor is the forecast horizon.

N is the number of cases used to compute all statistics for this horizon other than the MAPE (see below).

MAPE N is the number of cases used to compute the MAPEs for this horizon.This can be less than the total number of forecasts for the horizon when someof the actuals are zero, thus preventing computation of the MAPE.

MAD is the Mean Absolute Deviation for the horizon, over all series on thecurrent script line.

Cumulative MAD is the cumulative MAD for all horizons up to and includingthe current horizon.



82 Out-of-sample Testing

MAPE is the Mean Absolute Percentage Error for the horizon, over all serieson the current script line.

Cumulative MAPE is the cumulative MAPE for all horizons up to and including the current horizon.

GMRAE is the Geometric Mean of Relative Absolute Error for this horizon.

Below Upper Limit is the fraction of times that the actual was at or below theupper confidence limit as set in Settings Options. This allows you to calibratethe empirical upper limit to the theoretical upper limit. This is useful whenyou are using the upper limit to set stocking levels.

Figures listed in the row below the last forecast horizon refer to all horizons.Each group on the Navigator will produce such a report.

Comparing an Alternative Model

Right click Monthly on the Navigator to display the context menu. Select

Model>Very Simple Models>Simple Moving Average and set the number

of terms to 3. Then right click Monthly a second time and select Apply

Modifier(s) to>All items. We have now forecasted all items using a 3-monthsimple moving average.

Select the group Monthly on the Navigator and view the Out-of-Sample

Rolling Evaluation for Items in Group section of the Forecast Report view.



Comparing an Alternative Model 83

There are several ways to compare the results of this forecast and our expertselection forecast. The most common approach is to look at the MAPEs(Mean Absolute Percent Error) for the two. In our example, the MAPEs for

expert selection are lower than the MAPEs for the moving averages for everytime horizon—clear evidence that expert selection outperforms the movingaverages.

The other comparison statistic that can be used, is the GMRAE (GeometricMean of Relative Absolute Error). This statistic is calculated using the

relative error between the naive model and the currently selected model. AGMRAE of 0.55 indicates that the size of the current model’s error is only55% of the size of the error generated using the naive model for the same dataset. In our example the GMRAEs for expert selection are lower than theGMRAEs for the moving averages—further evidence that expert selection

outperforms the moving averages.




85

Lesson 10

Operating From the Command Line

When the command line to execute the program includes a project file(FPProj) and the \b parameter, Forecast Pro Unlimited will read the project,create the forecasts, save all output files and then exit. This feature is particularly useful when you are integrating forecasting with other softwaresystems and desire a “hands-off” approach.

Command-Line Operation

In this lesson we will first use Forecast Pro Unlimited interactively to create a project file. We will then exit the program and execute the script via thecommand line.

Start Forecast Pro Unlimited and select Settings>Options. Set the data modeto Xls, the forecast horizon to 12 and then click on the Command Line tab.



86 Operating From the Command Line

The Project Components to Apply section allows you to control how theforecasts are generated when the project is run from the command line.

If Modifiers is selected, the project’s forecast modifiers (if any are present)will be used to generate the forecasts in command-line mode. If this option isnot selected, all forecast modifiers in the project will be ignored when runningfrom the command line.

If Overrides is selected, the command-line run will begin by reading in thedata, generating the statistical forecasts and then it will apply any matchingoverrides/comments in the project to the new forecasts. If this option is notselected, all overrides in the project will be ignored when running from thecommand line.

If Settings is selected, the command-line run will use all of the settingsassociated with the project (e.g., all settings in the Settings dialog box,numeric output formatting, etc.). If this option is not selected, the defaultsettings (i.e., the settings that have been saved as the defaults for use with new projects) will be used when running from the command line.

The Output Files section allows you to specify the output files to generate

when the project is run from the command line. The format and content of these files will match the current project’s settings.

Make sure your settings match the ones above. Click the OK button to saveyour selections.

Click the red Script icon and define a script consisting of the single line

123 Bakery



Command-Line Operation 87

Click OK to accept the script and then select File>Save to save the project. Name the project “Test”. Exit Forecast Pro Unlimited.

We will now run the project from the command line.

Select Run from the Windows Start Menu. Execute the program includingTest.FPProj as a parameter followed by “ /b”. If you installed Forecast Pro

Unlimited into a directory named “C:\Program Files\Forecast Pro Unlimited v1” the command line should read (including the quotation marks):

"C:\Program Files\Forecast Pro Unlimited v6\ForecastProUnlimited.exe" Test.FPProj /b

After issuing the command, the program will execute the script and create allof the requested output files. Examine your output directory to confirm thefiles were created.

For more detailed information regarding command-line operation consult theReference section of this manual.




3

REFERENCE



91

Chapter 1

Setting Up Your Data

Forecast Pro Unlimited requires historic data to be input in a specific format.For many users, the data format is an initial stumbling block on the way tosuccess in forecasting their own data. This usually happens when a user misunderstands some of the details of the data format. This chapter explainsthe data formats supported by Forecast Pro Unlimited, advises you how toselect a format, and describes how to create the files. If you read the materialcarefully, and examine the sample data files that are shipped with Forecast ProUnlimited, you will be up and running quickly.

The first two sections describe data requirements and the different data fileformats. After you have selected your file format, consult the appropriatesection for instructions on how to create the files.

Overview

Forecast Pro Unlimited works with historic time series, i.e., sequences of values of a variable at some time interval. Business series are usuallyexpressed in terms of the calendar, and can be daily, weekly, monthly or quarterly. Forecast Pro Unlimited can work with any type of time series, but isoriented towards weekly, monthly and quarterly calendar data.



92 Setting Up Your Data

Collection and maintenance of reliable historic data is up to you and your ITdepartment, and can be implemented in many different ways, depending oninformation flow conditions in your organization.

Data length

Forecast Pro Unlimited works by fitting a statistical model to your historicdata and extrapolating it via the fitted model. Thus, your data must be longenough to provide reasonably stable estimates of the most important featuresof the data. Very short or very noisy historic records usually yield very simplemodels because the data are too short to support statistical estimates of important features like seasonality.

If the data are very short, say four points or fewer , Forecast Pro Unlimited can pick up neither seasonality nor trend, and reverts to the Simple MovingAverage model.

For more than four points but less than two years worth of data , Forecast Pro

Unlimited can fit and forecast trends but not seasonality. If your data are infact nonseasonal, your forecasts are likely to be adequate. If your data are infact seasonal, the forecasts are likely to be poor—Forecast Pro Unlimited cannot extract or forecast the seasonality and part of the seasonality may bemistaken for trending.

However, seasonal forecasts from short data sets are feasible when the shortdata sets are nested within aggregate product groups with longer histories. Theseasonality of the aggregate can be “borrowed” and applied to the short datasets. See the section entitled Multiple-level Scripts in the next chapter for details.

Seasonality can be estimated and forecasted from two to three years of data but this amount of data is marginal, especially when your data are noisy or intermittent. In some cases, patterns in the noise are mistaken for seasonality,yielding inappropriate “seasonal” forecasts.

Robust capture of seasonality requires three or more years of data. Four to

seven years is even better, since there is more information from which the program can separate seasonality and trend from the noise.



Overview 93

There is little additional payoff in accuracy beyond about seven years of data,and the cost in computer time can be substantial.

Missing values and zeros

Each of the statistical forecasting methods requires an unbroken stream of historic values. Unfortunately, however, missing values are common in business data. You therefore need a well-formulated approach to missing data.There are two parts to the approach. You must first decide how to encodemissing values in the input data file. Then you must decide how these missingvalues are to be treated by Forecast Pro Unlimited.

The encoding of missing values depends upon the data format. However, one problem is common to all of the formats—the distinction between missingvalues and zeroes. Forecast Pro Unlimited ordinarily considers zeroes to beactual data values and treats them as such. If they were really intended asmissing values, then the forecasts can be badly biased.

Forecast Pro Unlimited treats a sequence of leading zeroes as missing valuesif you check “Ignore leading zeroes” on the Formats tab of the

Settings>Options dialog box and it treats trailing zeroes as missing if youcheck “Ignore trailing zeroes”. If you can avoid using this strategy, do so. Itsdependence upon these flag settings makes it a little risky.

Zeroes in the middle of your data are always treated as numerical values.Therefore, embedded missing values must be explicitly coded as missing.Since the way you do this depends upon the data format, it will be addressed as we discuss each specific data format.

Forecast Pro Unlimited interprets your input data and missing values as thedata are read. Thus you cannot change treatment of missing data for a data setthat has already been read except by rereading it from scratch. Theinterpretation of zeros and missing values depends upon the “Ignore leadingzeroes” and “Ignore trailing zeroes” flags, and upon the “Missing values”

setting on the Formats tab of the Settings>Options dialog box. There arethree possible settings for missing values— Truncate, Impute or Zero.




Truncate directs Forecast Pro Unlimited to use only the most recent unbrokenstream of data. All data up to and including the last missing value arediscarded.

Impute directs Forecast Pro Unlimited to discard leading and trailing missingvalues but to impute embedded missing values via linear interpolation.

Zero directs Forecast Pro Unlimited to reset missing values to zeroes and totreat them as numerical values. This setting is appropriate when the data file iswritten from a database in which only the non-zero records are stored. In other cases it might lead to badly biased forecasts.

Header information

You will need to collect not only the time series data, but also certain header

information. The header information includes the following.

Variable name. The name by which the program refers to an individual item

(often an SKU). An item name is limited to 256 of the following characters:

0-9 A-Z a-z ! # $ % & ? @ _ + - / * . ~ < >

However, you may also be limited by restrictions imposed by the software towhich you export your data. SQL databases, for instance, sometimes requirethat the first character in a name is a letter, and do not regard upper and lower cases as distinct.

If the variable name begins or ends with an underbar (_), Forecast ProUnlimited tags it as a potential helper variable and will not forecast it. Helper

variables are used for event modeling and weighting transformations.

Variable description. A short description of the variable. Variabledescriptions must not exceed 256 characters in length. Variable descriptionsare displayed within Forecast Pro Unlimited and can be included in NumericOutput Files and Forecast Report Files.

Starting year and period . The year and period corresponding to the firstavailable data point. If the first available observation for a monthly series wasApril, 2001 then the starting year would be 2001 and the starting period would



Selecting a Data Format 95

be 4. The starting year must be 100 or later. For data that are not calendar-oriented, enter 100 for the starting year and 1 for the starting period.

Periods per year . 12 for monthly data, 4 for quarterly data, etc. For data thatis not calendar oriented, use 1. Note that most weekly corporate calendarsinvolve occasional years of 53 weeks—these occurrences will cause theforecast date tags to be off by a week.

Periods per cycle. The number of periods per seasonal cycle, usually equal tothe number of periods per year. An exception might be daily data, where periods per cycle could be set to 7 to capture weekly patterns and periods per year set to 365. Or, if weekend days are excluded, periods per cycle might be5 and periods per year 260. For nonseasonal data, periods per cycle should be1. This is a critical entry because Forecast Pro Unlimited uses it for seasonaladjustments.

Selecting a Data Format

The program accepts three different data formats—Text (MLT), Spreadsheet(XLS & WK*) and Open Database Connectivity (ODBC). Examples of eachfile type are shipped with the program.

Which data format you choose will depend on your application and databaseoperations. Before making your choice you should be aware of the followingadvantages and disadvantages of each file type.

Spreadsheet (XLS, XLSX & WK*) files

Uses Excel and Lotus format spreadsheets (XLS, XLSX, WK1, WK3, WK4)to hold data. You can make and forecast multiple spreadsheet files on onescript if you wish.

PROS. If you are comfortable using a spreadsheet then the spreadsheet formatallows you to create, update and manipulate your data in a familiar

environment. Processing time for xls files is faster than for ODBC. Creatingforecast reports from Excel is straightforward.




CONS. The layout of your spreadsheet must follow the Forecast ProUnlimited rules. You may have to change your current spreadsheet layout toone you find slightly less convenient.

Text (MLT) files

Uses text files to hold data. You can make and forecast multiple MLT files onone script if you wish.

PROS. Processing time is faster than for ODBC and slightly faster than for large spreadsheet files (1,000+ series). Most databases can output text files.

CONS. Editing, viewing and updating large files can be clumsy. Not alleditors accept very long files.

ODBC

Open database connectivity (ODBC) allows Forecast Pro Unlimited to read and write data directly to databases for which an ODBC driver exists. Most popular databases support ODBC including Access, Oracle and SQL server.

PROS. ODBC can provide direct access to data stored in the corporatedatabase obviating the need for intermediate files.

CONS. Reading and writing directly to the corporate database can raisesecurity issues. Data transfer can be slow for some configurations.

The next three sections supply the details you need to set up your data in oneof the Forecast Pro Unlimited formats. If you have chosen a format, you need

read only the pertinent section.

Spreadsheet (XLS, XLSX andWK*) Formats

Forecast Pro Unlimited can read Excel (XLS, XLSX) files and Lotus (WK1,WK3 and WK4) files.



Spreadsheet (XLS, XLSX and WK*) Formats 97

You must create your spreadsheet file from within your spreadsheet program.Forecast Pro Unlimited checks data types as it reads the spreadsheet. If it seestext where it expects a number, or a floating point number where it expects aninteger, an error message will be displayed. A blank cell is interpreted as a

missing value. Do not use zeros to represent missing values ⎯ they will beinterpreted as numbers and will probably distort your forecasts badly.

Entire spreadsheet vs. named range

You may either devote an entire worksheet to the data (recommended) or specify a portion of a worksheet by giving it the range name BFSDATA.

In a multi-worksheet workbook the data are assumed to reside in the first(topmost) worksheet unless an alternate sheet has been specified by naming itBFSDATA.

The cell references in the examples below assume that the entire worksheethas been devoted to the Forecast Pro Unlimited database. If you opt to specify

a named range, cell A1 in the samples will correspond to the cell in theuppermost left corner of your named range.

Row format

If your data are already stored by rows, you will want to consider the rowformat first. You can probably alter your spreadsheet to the Forecast Pro

Unlimited row layout in just a few minutes. In row format each time seriesoccupies a single row on the spreadsheet.

The data are assumed to reside in the topmost (first) spreadsheet in the

workbook unless an alternate sheet has been specified by naming itBFSDATA.

Row 1 of the spreadsheet is devoted to column headings. The keyword

“description” must be used as the column heading for the description column.The other cells in row 1 are ignored by Forecast Pro Unlimited. You can usethem for titles, calendar information, etc.

Here is an example of a row format spreadsheet.



Spreadsheet (XLS, XLSX and WK*) Formats 99

Column format

If your data are already stored by columns, you will want to consider thecolumn format first. You can probably alter your spreadsheet to the ForecastPro Unlimited row layout in just a few minutes. In column format each timeseries occupies a single column on the spreadsheet.

The data are assumed to reside in the topmost (first) spreadsheet in theworkbook unless an alternate sheet has been specified by naming itBFSDATA.

Column A of the spreadsheet is devoted to row headings. The keyword “description” must be used as the row heading for the description row. Theother cells in column A are ignored by Forecast Pro Unlimited. You can usethem for titles, calendar information, etc.

Here is an example of a column format spreadsheet.

The above example does not use attribute fields to define product groupings.This is often desirable and is discussed in the Using attribute fields to define a

hierarchy section.




In our example, each data record consists of six header items in rows 1through 6, followed by the historic data in remainder of the rows, beginning inrow 7. Each item is placed in a separate cell.

Rows 1 through 6 contain the following six required items.

1 Variable name. Up to 256 characters.

2 Variable description. Up to 256 characters.

3 Starting year. Must be an integer.

4 Starting period. Must be an integer.

5 Periods per year. Must be an integer.

6 Periods per seasonal cycle. Must be an integer.

See the Overview section at the beginning of this chapter for more details onthese items.

The time series data begin in row 7. The starting year (row 3) and starting period (row 4) refer to the year and period for row 7, whether or not that cellactually contains data. Thus in the example spreadsheet, the first value for PG1-204-2 is for January 2003 in row 9, even though the header defines thestarting period as 11 and the starting year as 2002.

Formatting rules

Keep the following points in mind when you are creating your spreadsheet.

If you are not using attribute fields each variable name must be unique.

If you are using attribute fields each variable name within a given group must be unique.

If you use zeroes to pad the data prior to an item's availability, make sure thatIgnore leading zeros is selected on the Formats tab of the Settings Optionsdialog box.



Text (MLT) Formats 101

Text (MLT) Formats

A multivariate text file contains information for all the series you wish toforecast from a single line of the script file.

This file is stored as an editable text file with the extension MLT. You mustcreate this file outside of the program, making sure that you format itappropriately.

The MLT file format is straightforward. The file begins with record headingsfollowed by a semicolon. The record headings are required for the Variable Name and Description fields and are optional for all other records. You canuse them for titles, calendar information, etc. if you wish.

After the record headings comes header information for the first variable,followed by decimal numbers representing the historical data and then asemicolon. This is followed by the same format for subsequent time series.Two semicolons in a row mark the end of the MLT file.

Since Forecast Pro Unlimited recognizes spaces, tabs and returns asequivalent terminators, you have a great deal of flexibility in formatting your data. The most common format is to organize series by rows, as in thefollowing example.





hierarchy section below.

The header consists of the first six items:

Variable name. Up to 256 characters enclosed in quotes.

Variable description. Up to 256 characters enclosed in quotes.

Starting year. Must be an integer.

Starting period. Must be an integer.

Periods per year. Must be an integer.

Periods per seasonal cycle. Must be an integer.


The remaining fields contain historic data. Each time series is terminated witha semicolon, and the file is terminated with two semicolons.

The one-time-series-per-line format illustrated above is easy to read, however,keep in mind you can use any combination of rows and columns.

Formatting rules

Keep the following points in mind when you are creating your .MLT files.

The record headers, variable name field and variable description field must

always appear in quotes.

If you are not using attribute fields, each variable name must be unique. If youare using attribute fields, each variable name within a given group must beunique.

If you plan to use commas to separate items, or as place holders for largenumbers, you will need to set a few options on the format tab of the SettingsOptions dialog box to insure that they are interpreted correctly.



Using ODBC 103

Be careful not to use dollar signs, page breaks or other symbols that mayconfuse Forecast Pro Unlimited.

Use the word MISSING instead of a number to indicate a missing value.

If you use zeroes to pad the data prior to an item's availability, make sure thatIgnore leading zeros is selected on the Formats tab of the Settings Options

dialog box.

Using ODBC

ODBC provides direct data communications between Forecast Pro Unlimited and a wide variety of databases. This is accomplished through intermediaryODBC drivers that lie between Forecast Pro Unlimited and your database.You must obtain the driver from the database manufacturer or a third partyand install it according to the directions provided.

ODBC drivers are available for many database products including Access,Oracle, DB2, SQL Server, FoxPro, Paradox and Btrieve.

Database structure

Forecast Pro Unlimited reads data from structured tables or views defined inthe database. Ordinarily, the tables created for the Forecast Pro Unlimited interface are just a subset of the entire database.

Here is an example of a of a data table formatted for Forecast Pro Unlimited.





hierarchy section.

In our example, each data record consists of six header items in columns 1through 6, followed by the corresponding historic data point. It is important tonote that each data record must contain the aggregated value for the given period.

The fields are defined as follows.

ItemId0. This is a text field containing the variable name. It can be up to 256characters.

Description. This is a text field containing the variable description. It can beup to 256 characters.

Hist_Year . This is a number field with field size double containing the year.

Hist_Period . This is a number field with field size double containing the period.

Ppy. This is a number field with field size double containing the periods per year.



Using ODBC 105

Ppc. This is a number field with field size double containing the periods per cycle.

Hist_Value. This is a number field with field size double containing theaggregated historic data value for the record’s “date” as defined by theHist_Year and Hist_Period.


To connect to an ODBC database you set the data mode to ODBC in Settings,

and then select Operations>ODBC Connect. You will be prompted toestablish the file data source (select an ODBC driver) and then to select adatabase.

When you invoke the Define Script dialog box, the list box to the rightincludes all tables from your database that do not end with an underbar ( _ ),including those that have nothing to do with Forecast Pro Unlimited. A table

from the database is treated just like a file when you are using one of the filedata modes. Thus you should arrange data into the tables or views that makelogical sense as entries in the script.

Formatting rules

ODBC drivers vary a great deal. If you experience any difficulties connectingto the database, the first thing to check is that you are using the latest ODBCdriver available for your database.

With the exception of the last one (which only pertains to Oracle) all of the

notes below have been verified using Microsoft Access. There is a chance thatyour ODBC driver might be more (or less) restrictive.

Missing values are indicated by whatever rules are used by the nativedatabase. Every database recognizes the distinction between zeroes and missing values. If you want, you can encode leading and trailing missingvalues as zeroes. In that case you must make sure that “Ignore leading zeros”

and/or “Ignore trailing zeroes” is selected on the Format tab of the Settings

Options dialog box.




Table names and variable names should begin with a letter and should notinclude blank spaces.

If you are not using attribute fields each variable name must be unique.

If you are using attribute fields each variable name within a given group must be unique.

Oracle Users: Some Oracle drivers will only work if the Hist_Value field isdefined as FLOAT (not NUMBER). All other numeric fields can be NUMBER with the decimal places set to zero.

Using Attribute Fields to Define aHierarchy

If you do not include any attribute fields, when you read the data into ForecastPro Unlimited, it will create a one-level hierarchy (i.e., there will be no group

totals).

If your data file, table or query includes attribute fields prior to the Variable Name field, Forecast Pro Unlimited will automatically build a multiple-levelhierarchy. The first level of the hierarchy will correspond to the first attributefield, the second level of the forecasting hierarchy will correspond to the

second attribute field, etc.

Let’s illustrate this with an example. The spreadsheet 123 Bakery.xls is asample data file shipped with Forecast Pro Unlimited. It is shown below.



Using Attribu te Fields to Define a Hierarchy 107

In our example, our first attribute field is Total. Notice that the value for thefield Total is “Total” for all items. Since all items have the same value for thefirst attribute field, when we read this spreadsheet into Forecast Pro Unlimited

the first level of our forecasting hierarchy will provide a grand total for allitems in our spreadsheet. Notice that the next attribute field is Category,which contains both Muffins and Cakes. Thus, when we read this spreadsheet

into Forecast Pro Unlimited the second level of our forecasting hierarchy will breakdown total sales into Muffins and Cakes. The next attribute field isCustomer. Thus, the third level of our forecasting hierarchy will breakdownthe Muffins and Cakes into Customers. The next field is the Variable Name(labeled “SKU”) so the final level of our hierarchy will breakdown theCategory-by-Customer into SKUs.

The order of appearance on the spreadsheet dictates the structure of the

hierarchy so you’ll want to take care when constructing your spreadsheet.Often this entails sorting your spreadsheet on the attribute fields prior toreading it into Forecast Pro Unlimited.

If you use a pivot tables in Excel to create your input files the resultingspreadsheet will only include records in the attribute fields when the attributeschange. Thus, the data in our example would be represented as shown below.




Forecast Pro Unlimited will read spreadsheets formatted in this fashion properly.

ODBC attr ibute fields

If you are using ODBC the first attribute field must be named ItemId0, thesecond attribute field must be named ItemId1, etc. All attribute fields must betext fields.



109

Chapter 2

Saving Your Work

Forecast Pro Unlimited can save five different types of output.

Forecast Projects allow you to save your forecasting session so that you canreturn later and pick up where you left off or to share the session with others.

The forecast project saves the data, forecasts, overrides, Hot List, modifiersetc. as they currently exist in memory. If you update your input data files after saving a project, when you reopen the project you’ll have the option of either generating new forecasts using the updated data or restoring the conditions present when the project was saved (i.e., not recalculating the forecasts usingthe new data).

Numeric Output Files can contain time series output such as history, forecasts,

fitted values and confidence limits as well as statistical output such as theforecasting model specification and summary statistics. These files can besaved in text, spreadsheet, ODBC or XML formats. You can include outputfor either all items forecasted or just for the Hot List. Generally speaking, if you will be importing the forecasts into another application these are the filesyou will want to use.

Formatted Forecast Reports can be saved to Excel. You can save thesereports for the currently displayed item, all items on the Hot List or all itemsforecasted. The Excel workbook will include a separate worksheet for each



110 Saving Your Work

item containing a graph (optional) and the on-screen Forecast Report. Thesereports are convenient if you wish to present the forecasts and/or models used to colleagues.

The Override Report can be viewed on-screen and saved to Excel. This reportlists items where overrides and/or comments have been made.

The Outlier Report can be viewed on-screen and saved to Excel. This reportslists items where outliers have been detected and/or corrected.

The following sections discuss these output files in more detail.

Forecast Projects

Forecast Projects allow you to save your forecasting session so that you canreturn later and pick up where you left off or to share the session with others.The forecast project saves the data, forecasts, overrides, Hot List, modifiersetc. as they currently exist in memory. If you update your input data files after saving a project, when you reopen the project you’ll have the option of either

generating new forecasts using the updated data or restoring the conditions present when the project was saved (i.e., not recalculating the forecasts usingthe new data).

Opening Projects

To save a forecast project you select File>Save or click the blue Save Project

icon ( ). To open a forecast project you select File>Open or click the blue

Open Project icon ( ).

If you open a project and data files have not changed (i.e., the time and datestamps match), the project will restore the conditions present when the projectwas saved allowing you to continue your work.

If you open a project and the data files are not available, the project willrestore the conditions present when the project was saved—thus you canreview projects, make adjustments, etc. even if you do not have access to thedata files. This will often be the case when sharing projects with colleagues.



Numeric Output Files 111

If you open a project and data files have changed, a dialog box will appear allowing you to either generate new forecasts using the updated data or restorethe conditions present when the project was saved (i.e., do not recalculate the

forecasts using the new data).

Project Files

The project consists of three separate data files.

The Project Settings File (*.FPProj) is an XML file containing the projectsettings and script definition.

The Project Tables File (*.mdb) is a Microsoft Access database filecontaining separate tables for the in-place overrides/comments, forecastmodifiers and Hot List.

The Project Snapshot File (*.fcb) is a binary file that saves the forecastcomponents that are currently in memory (e.g., the forecasts, summarystatistics, etc.) so that the session can be restored without having to read in thedata, recalculate the forecasts, etc. This allows you to return to your sessionwhere you left off and to share the session with others.

When sharing projects with others, you need to provide all three files.

Numeric Output Files

Numeric Output Files can contain time series output such as history, forecasts,fitted values and confidence limits as well as statistical output such as theforecasting model specification and summary statistics. These files can be




saved in text, spreadsheet, ODBC or XML formats. You can include outputfor either all items forecasted or just for the Hot List. Generally speaking, if you will be importing the forecasts into another application these are the filesyou will want to use.

You can specify the format and content of Numeric Output File using

Settings>Numeric Output Design. The operation of the Numeric Output

Design dialog box is described in the Command Reference section of thismanual.

You can preview the contents of the current Numeric Output File using the

Numeric Output view. You can open this view using View>Numeric Output or using the yellow Preview Numeric Output icon ( ).

To save a Numeric Output file for all items forecasted, select

Project>Export>Full Numeric Output or click the purple Save NumericOutput icon ( ).

To save a Numeric Output file for the current Hot List you selectProject>Export>Hot List Numeric Output or use the Save Numeric

Output option on the Hot List’s context menu.

Formatted Forecast Reports

Formatted Forecast Reports can be saved to Excel. You can save thesereports for the currently displayed item, all items on the Hot List or all itemsforecasted. The Excel workbook will include a separate worksheet for eachitem. These reports are convenient if you wish to present the forecasts and/or

models used to colleagues.

Each report contains the information found in the current Forecast Reportview and (optionally) the graph found in the current graph view.

You can specify the format and content of Formatted Forecast Report using

Settings>Forecast Report Design and Settings>Graph Settings. Theoperation of the Forecast Report Design and Graph Settings dialog boxes aredescribed in the Command Reference section of this manual.



Override Reports 113

To save a Forecast Report file for all items forecasted, you select

Project>Export>Full Forecast Report or click the purple Save Formatted Forecast Report icon ( ).

To save a Forecast Report file for the current Hot List you select

Project>Export>Hot List Forecast Report or use the Save Forecast

Reports option on the Hot List’s context menu.

You can print a forecast report for the currently displayed item using

File>Print, the blue Print icon or the Print this page option on the ForecastReport view’s context menu. You can also save an Excel report for the

currently displayed item using the Save this page to Excel option on theForecast Report view’s context menu.

Override Reports

Override Reports can be viewed on-screen and saved to Excel. These reports

list the items where overrides and/or comments have been made.

To open an Override Report you select View>Override Report or click theyellow View Override Report icon ( ). To save an Override Report, select

Project>Export>Override Report or use the view’s context menu.

You can specify the format and content of the report using

Settings>Override Report Design. The operation of the Override ReportDesign dialog box is described in the Command Reference section of thismanual.

Outlier Reports

Outlier Reports can be viewed on-screen and saved to Excel. These reportslist the items where outliers have been detected and/or corrected.

To open an Outlier Report you select View>Outlier Report or click theyellow View Outlier Report icon ( ). To save an Outlier Report you select

Project>Export>Outlier Report or use the view’s context menu.




The content and format of the report is controlled using the Outlier Report’scontext menu.



115

Chapter 3

Working with Scripts and Hierarchies

The script is a list of the data files to be forecasted. Together, the script and the data files define the hierarchy that will appear in the Navigator when youread the data into Forecast Pro Unlimited.

This chapter explains how to define scripts and also discusses the strategiesthat Forecast Pro Unlimited uses to produce consistent forecasts for hierarchical data.

Defining the Script

The information contained in the script is recorded in the current forecast

project. In the case of a new forecast project, the script is initially blank. If you open an existing forecast project, the script is the same as when the project was saved.

The Define Script dialog box pictured below is accessed by clicking on the

red Script icon ( ) or by selecting Operations>Script.



116 Working with Scripts and Hierarchies

The dialog box consists of the script area and the data list box. You can usethe vertical scroll bar to see other lines of the script.

The data list box displays the currently available data files or ODBC tables. If you double-click a filename or tablename in the box, the text is transferred tothe currently selected script line.

Defining the Hierarchy

Product data can almost always be organized into several levels of aggregation. Suppose that an SKU (Stock Keeping Unit) consists of a size and color of a shoe in a certain style. These are the items of importance for production and inventory control. The corporation might aggregate theseSKU's first into styles and then into style lines for marketing and sales. Thesemight be aggregated further into geographical regions for the benefit of topmanagement.

Forecast Pro Unlimited allows you to define product hierarchies and createone set of self-consistent forecasts. It allows top-down or bottom-upreconciliation, seasonal adjustment based upon aggregate data and model

selection at the aggregate level.

It is not necessary that item histories begin and end at the same time. Thusitems may consist of SKU's that have been retired or replaced by new SKU's.

Obsolete SKU's will contribute to the group level history but will notthemselves be forecasted. You will notice that the starting and ending datesfor the overall script consist of the starting date for the oldest item and theending date for the newest. Forecasts will be prepared for all items and groupsthat are “alive” at the end of the data set. Those whose histories terminate before that time are considered dead—they contribute to the historic



Defining the Hierarchy 117

aggregates (and therefore influence aggregate forecasts) but they are notthemselves forecasted.

Forecast Pro Unlimited allows you to organize your data into hierarchies twodifferent ways. The most common approach is to define the groupings directlyin the data file using attribute fields. This approach was discussed in theSetting Up Your Data chapter.

The other approach is to define the groups on your script. Here, for example,is a script that a beer manufacturer might use

This script defines three aggregate groups named Brand, Retail and Commercial. These group variables are not in the database—they are created internally by aggregating all data files found between a _GROUP= statementand its corresponding _END flag.

Thus, the group Retail is defined as the sum of all of the time series found inthe data files Cans and Bottles. The group Commercial is defined as the sumof all of the time series found in the data file Kegs. The group Brand aggregates the two lower-level groups Retail and Commercial. Notice thateach group definition begins with a _GROUP=NAME line and ends with an

_END. The indentation is optional and serves only to make the script easier toread.

In addition to using the _GROUP= and _END modifiers on the script you canalso define groups within a data file using attribute fields. Groups defined within files will appear on the Navigator when the data are read in. You cancombine script-based and within-file group definitions if you wish.




Reconci ling the Hierarchy

If no group-level reconciliation modifiers have been specified, a bottom-upapproach will be used to reconcile the forecasts. The procedure operates asfollows.

First Forecast Pro Unlimited prepares forecasts for each and every group and item.

Then it recomputes the group-level forecasts by aggregating the constituentforecast (bottom up). The original group-level forecasts are replaced, but thewidth of their confidence limits are retained and re-centered on the newforecasts.

Two group-level modifiers can be used to specify alternative reconciliationapproaches. These modifiers can only be used on group-level data.

\TOPDOWN directs Forecast Pro Unlimited to perform top-down

reconciliation for all members of the indicated group.

This procedure begins by preparing forecasts for each and every group and item.

Then it proportionally adjusts the nested lower level items and groupsforecasts to sum to the \TOPDOWN group forecast. If the \TOPDOWN groupis itself nested within larger groups, the forecasts for the larger groups arecomputed by the bottom-up approach. (An example of top-downreconciliation is given below.)

You cannot define a \TOPDOWN group as a member of another \TOPDOWNgroup.

\INDEXES directs Forecast Pro Unlimited to calculate the seasonal indexes atthe indicated group level and use them to deseasonalize the histories for thenested items and groups. The deseasonalized series are then forecasted and asa final step reseasonalized. This approach works well when the items sharethe same seasonality. It allows one to deal with items whose histories are tooshort to extract seasonality directly. You cannot nest \INDEXES groups



Reconcil ing the Hierarchy 119

within each other. Model parameters are reestimated for each item of the\INDEXES group.

The following screenshot illustrates the use of the \TOPDOWN modifier.

In the example above, \TOPDOWN directs Forecast Pro Unlimited tocompute forecasts and confidence limits for the aggregate groups Retail and Commercial at the aggregate level. The item level forecasts and confidencelimits are multiplicatively adjusted so that the item level forecasts sumcorrectly to the group level forecasts. Forecasts for Brand are obtained by bottom-up aggregation of the groups Retail and Commercial.

Note the assumptions behind this approach. We have assumed that the nested items are statistically similar, so that we can obtain better forecasts of the

groups Retail and Commercial by forecasting them as a unit. However, theforecast models for these two groups are judged as distinct compared to theirregularity. Thus the best forecasts of Brand are obtained by directaggregation of the two member groups.

It would be illegal to add the modifier \TOPDOWN to the line definingBRAND unless we remove the \TOPDOWN modifiers from the two member groups. In this case the entire hierarchy would be reconciled using top-downaggregation. Usually, in a hierarchy involving three or more levels of aggregation, you will want to place the \TOPDOWN modifiers somewherenear the middle.




Two things happen as we move from item level data upwards through thehierarchy.

The effects of irregularity decrease through aggregation.

Statistical models become more complex and distinct from each other. Thusthe signal-to-noise ratio increases as we move upwards through the hierarchy.

Now consider the case where both the \TOPDOWN and \INDEXES keywordsare used for Retail and Commercial. In this case, Forecast Pro Unlimited extracts seasonal indexes for these groups at the group level. It thendeseasonalizes each item level history by using the indexes from the parentgroup, forecasts the resulting nonseasonal data, and then reseasonalizes theforecasts. The forecasts are then reconciled as already described.

A note about negative values

In a multiple-level problem, it is assumed that the data are basically

nonnegative—the aggregate level data must be non-negative, but the itemlevel data can include a few negatives.

Negative sales figures are relatively common since many companies register returns as negative sales. Furthermore, returns are often accumulated on the books and taken as occasional “hits” on sales. While this may make good accounting sense, it raises forecasting problems and will certainly decreaseaccuracy. Ideally, companies would either distribute returns to the months inwhich the products were shipped, or maintain returns as an entirely differenthistoric record. Nevertheless, listing sales as negative is common and must bedealt with in some way.

Beyond an expected decrease in accuracy for forecasts from a product withnegative sales, negatives pose a technical problem for product hierarchies.Multiplicative seasonal indexes cannot be extracted from a nonpositive seriesand additive indexes cannot be disaggregated to nested products. Thus, if thegroup level data contain negatives, Forecast Pro Unlimited will use anadditive index model and seasonal disaggregation will not be allowed (i.e., the\INDEXES flag will be ignored). In the case of positive group level data withnegative values in the constituent data, the multiplicative adjustment of



Reconcil ing the Hierarchy 121

negative values may have a different effect than one expects—a seasonalindex of 1.5 to a sales figure -100 yields the value -150.

Top-down adjustment is also problematic when negative forecast values areinvolved. Therefore, Forecast Pro Unlimited clips negative forecasts to zero if you have specified top-down disaggregation, regardless of how you have set

Allow negative forecasts in the Options dialog box in the Settings menu.



123

Chapter 4

Using Forecast Modif iers

By default, Forecast Pro Unlimited will automatically select a forecastingmodel for each item on the Navigator using expert selection. The expertselection option works extremely well and is the method of choice for themajority of Forecast Pro Unlimited users. The expert selection algorithm isdescribed in the Forecast Pro Statistical Reference Manual.

Alternatively you can dictate the models and/or forecasting options using forecast modifiers. Forecast modifiers are added to items on the Navigator using the Navigator’s context menu.

This chapter documents the available modifiers. A complete listing of theavailable modifiers is also found in the on-line help system.

Model Specification Modifiers

Model specification modifiers are used to dictate that a specific model be used

for the time series.

Supported modifiers are listed below grouped by model type.



124 Using Forecast Modif iers

Expert Selection

Model>Expert Selection (no modifier). If an item on the Navigator does notcontain a modifier then Forecast Pro will use its expert selection algorithm toautomatically select the appropriate forecasting method. If an item on the Navigator does contain one or more modifiers and you select Model>Expert

Selection the modifiers will be removed and Forecast Pro will use its expert

selection algorithm to automatically select the appropriate forecasting method.

Exponential Smoothing

Model>Exponential Smoothing>Auto: \EXSM. Use the automatic fittingexponential smoothing model.

Model>Exponential Smoothing>Simple: \SIMPLE. Use the simpleexponential smoothing model.

Model>Exponential Smoothing>Holt: \HOLT. Use the Holt exponential

smoothing model.

Model>Exponential Smoothing>Winters: \WINTERS. Use the Wintersexponential smoothing model.

Model>Exponential Smoothing>Custom: \EXSM=XY. Use a customexponential smoothing model with trend type X (N=no trend, L=linear trend,D=damped trend, E=exponential, *=Forecast Pro decides), seasonality type Y(N=nonseasonal, M=multiplicative seasonal, A=additive seasonal, *=ForecastPro decides) and optimized smoothing weights.




negative binomial). This option is used to obtain better estimates of theconfidence limits for low volume integer series (typically with many zeros).

Intermittent Data

Model>Intermittent: \INTER. Use the Croston’s intermittent data model.

Very Simple Models

Several of the menu options available under Model>Very Simple Models

invoke the dialog box below.

Model>Very Simple Models>Simple Moving Average: \SMA=n. Use an n-term simple moving average. If the Automatic option is selected Forecast ProUnlimited will choose the number of terms to use and the forecast modifier will be displayed as \SMA.

Model>Very Simple Models>Same as Last Year>No Change: \SALY. Seteach forecasted value to equal the value for the same period last year (either

history or forecast as the case may be).

Model>Very Simple Models>Same as Last Year>Plus Percent: \SALYP=n. Set each forecasted value to equal the value for the same period last year plus the specified percentage (n).

Model>Very Simple Models>Same as Last Year>Plus Increment: \SALYI=n. Set each forecasted value to equal the value for the same period last year plus the specified increment (n).



Model Specification Modifiers 127

Model>Very Simple Models>Same as Last Year>Plus Delta Percent: \SALYDP. Set each forecasted value to equal the value for the same period lastyear adjusted by the percentage change between the value for the same period last year and the same period two years ago.

Model>Very Simple Models>Same as Last Year>Plus Delta Increment: \SALYDI. Set each forecasted value to equal the value for the same period last

year adjusted by the unit change between the value for the same period lastyear and the same period two years ago.

Model>Very Simple Models>Fixed Forecast Value: \FIXED=n. Set eachforecasted value to n.

Model>Very Simple Models>Fixed Forecast Value: \FIXEDAFT=(Year,

Period, n). Set each forecasted value which falls after the year and period specified to n.

Curve Fitting

Models>Curve Fit>Automatic: \CFIT. Use the curve fitting routine thatminimizes the BIC over the historic data.

Models>Curve Fit>Straight Line: \CFIT=LINE. Fit a straight line to the

data set.

Models>Curve Fit>Quadratic: \CFIT=QUAD. Fit a quadratic curve to thedata set.

Models>Curve Fit>Exponential: \CFIT=EXPO. Fit an exponential curve tothe data set.

Models>Curve Fit>Growth Curve: \CFIT=GROW. Fit a growth curve tothe data set.




Event Model Modif iers

To build an event model, you must tell the program when events of each typeoccur. To do this, you must construct an event schedule which classifies each period by event type (0=no event, 1=event of type 1, 2=event of type 2, etc.).The format is the same as that for any other historic data record, except that its

entries are all small integers. The event schedule must be defined for each period in the historic record. If you want to forecast the effects of futureknown events, you must include these future periods as well.

The event schedule is a helper variable. Helper variables names must start or end with an underbar. Helper variables are not forecasted and their values arenot included in group totals. Helper variables are used solely in conjunctionwith event models and weighting transformations.

For examples of constructing event variables and building event models,consult Building Event Models in the tutorial section. For more statisticaldetails, consult the section on exponential smoothing in the Forecast Pro

Statistical Reference Manual.

The following modifiers are associated with event models.

Events>Select: \EVENT=_X. Use an event model. _X is the name of the timeseries containing the event schedule.

Model>Exponential Smoothing>Custom: \EXSM=XYZ. Use an exponentialsmoothing model with trend type X (N=no trend, L=linear trend, D=damped trend, E=exponential, *=Forecast Pro decides), seasonality type Y(N=nonseasonal, M=multiplicative seasonal, A=additive seasonal, *=Forecast

Pro decides) and event type Z (M=multiplicative, A=additive, *=Forecast Prodecides). This modifier can only be used in conjunction with \EVENT=.

Model>Exponential Smoothing>Custom: \XYZ(A,B,C,D,E). Use anexponential smoothing model with trend type X (N=no trend, L=linear trend,D=damped trend, E=exponential, *=Forecast Pro decides), seasonality type Y(N=nonseasonal, M=multiplicative seasonal, A=additive seasonal, *=ForecastPro decides), event type Z (M=multiplicative, A=additive, *=Forecast Pro

decides) and user defined smoothing weights (A=level, B=trend,



The Weighting Transformation Modifier 129

C=damping/growth, D=seasonal, E=event). This modifier can only be used inconjunction with \EVENT=.

The Weighting TransformationModifier

The weighting transformation is most commonly used to deseasonalize your variables using externally supplied seasonal weights, or to normalize the datafor trading day effects (e.g., 4-4-5 calendars, number of working days per month, etc.).

To use the weighting transformation you must create a helper variablecontaining the weights. Helper variables names must start or end with anunderbar. Helper variables are not forecasted and their values are not included in group totals. Helper variables are used solely in conjunction with eventmodels and weighting transformations.

The following modifier is used to specify the weighting transformation.

Weights>Select: \WGT=_X. Use an weighting transformation. _X is the nameof the time series containing the weights.

The procedure divides each value of the specified time series by the

corresponding value (weight) in _X. It then forecasts the deweighted variableand multiplies the forecasts by their corresponding weights. The weightingvariable _X must span the entire history and forecast period for each variableto be forecasted.

Outlier Detection/CorrectionModifiers

When you generate forecasts, Forecast Pro Unlimited uses the current settingsin the Outliers tab to determine the default behavior for detection and correction or outliers. There are three choices:

None. Do not detect nor correct for outliers.




Detection only. Detect and report outliers but do not correct for them (i.e., base the forecasts on the uncorrected values).

Detection and correction. Detect and correct outliers (i.e., base the forecastson the corrected values).

You can override this project level-setting for individual items using the

forecast modifiers listed below. Forecast Pro Unlimited only detects outliersfor end items (i.e., non-group level data), so the modifiers cannot be used for groups.

Outliers>Default: (no modifier). Revert back to the default project-levelsetting for the specified item. This option is used to remove any of the outlier modifiers (listed below) from the Navigator.

Outliers>Off : \OUTLIER=OFF. Do not detect nor correct outliers for thespecified item.

Outliers>Detect: \OUTLIER=DETECT. Detect and report outliers for thespecified item, but do not correct for them (i.e., base the forecasts on theuncorrected values).

Outliers>Correct: \OUTLIER=CORRECT. Detect and correct outliers for thespecified item (i.e., base the forecasts on the corrected values).

Reconci liation Modifiers

If no group-level reconciliation modifiers have been specified, a bottom-upapproach will be used to reconcile the forecasts. The procedure operates as

follows.

First Forecast Pro Unlimited prepares forecasts for each and every group and item.

Then it recomputes the group-level forecasts by aggregating the constituentforecast (bottom up). The original group level forecasts are replaced, but their confidence limits are retained and proportionately adjusted.



Miscellaneous Custom Modifiers 131

Two group-level modifiers can be used to specify alternative reconciliationapproaches. These modifiers can only be used on group-level data.

Top down: \TOPDOWN. directs Forecast Pro Unlimited to perform top-downreconciliation for all members of the indicated group.

This procedure begins by preparing forecasts for each and every group and

item.

Then it proportionally adjusts the nested lower level items and groupsforecasts to sum to the \TOPDOWN group forecast. If the \TOPDOWN groupis itself nested within larger groups, the forecasts for the larger groups arecomputed using a bottom-up approach.

Indexes: \INDEXES. directs Forecast Pro Unlimited to calculate the seasonalindexes at the indicated group level and use them to deseasonalize thehistories for the nested items and groups. The deseasonalized series are thenforecasted and as a final step reseasonalized. This approach works well when

the items share the same seasonality. It allows one to deal with items whosehistories are too short to extract seasonality directly. You cannot nest\INDEXES groups within each other. Model parameters are reestimated for each item of the \INDEXES group.

Miscellaneous Custom Modifiers

The modifiers listed below can be entered from the keyboard via Custom

Modifier on the Navigator’s context menu.

Custom Modifier: \LOWER=l. Set the lower confidence limit equal to thevalue l. L must be between 0.1 and 50.0 inclusive.

Custom Modifier: \UPPER=u. Set the upper confidence limit equal to thevalue u. U must be between 50.0 and 99.9 inclusive.

Custom Modifier: \CONF(l,u). Set the lower confidence limit equal to thevalue l and the upper confidence limit equal to u. l must be between 0.1 and 50.0 inclusive and u must be between 50.0 and 99.9 inclusive.




Custom Modifier: \LEADTIME=n. Set the lead time for the safety stock to n.The specified lead time appears highlighted in the Safety Stock section of theForecast Report and determines the values written to the numeric output file.

Custom Modifier: \POW=key. Use a Box-Cox power transformation. Keymay equal LOG, SQRT, INVERSE or any decimal number from -1.0 to 1.0. If you enter \POW=AUTO, Forecast Pro Unlimited will test your data and

implement an appropriate Box-Cox transformation (or none at all). Consultthe Forecast Pro Statistical Reference Manual for a discussion of the Box-Cox power transformations.



133

Chapter 5

Command Reference

This chapter presents a command overview, a description of the user interface,a detailed description of each menu command and instructions on how todrive the program from the command line.

Operations Overview

To prepare forecasts from Forecast Pro Unlimited, you must follow thefollowing procedure.

Prepare the database.

Specify the data format and output options. This is accomplished using theSettings menu. Options set by the user will apply to the current forecastdocument and can also be saved to the ForecastProUnlimited.ini file and used as defaults for all future forecast documents using the “Set as Default” button.Thus, you need only set the options when you use Forecast Pro Unlimited for

the first time, or when you wish to change the current settings.

Create the script. The script is a list of the data files to be forecasted. It iscreated using the Define Script dialog box, which can be accessed from

Operations>Script or via the red Script icon ( ).



134 Command Reference

Select Operations>Read Data. Forecast Pro Unlimited will then read thedata from disk into RAM and display the starting and ending dates for theforecast run on the dialog bar.

Select Operations>Forecast. The program will then execute the script to prepare the forecasts. The forecasts are retained in memory and support thefollowing optional steps.

Select View>Overrides to adjust the forecasts and, if appropriate, reconcilethem across multiple levels.

Select View>Forecast Report to obtain text reports about individualforecasts.

Select View>Graph to view the forecasts graphically.

Select Project>Export>Full Numeric Output to save them to disk or to your database via ODBC.

The remaining sections of this chapter document this procedure, as well as allother options, in more detail.

The Forecast Pro UnlimitedInterface

This section describes the Forecast Pro Unlimited interface.



The Forecast Pro Unlimi ted Interface 135

menu bar Æ

toolbar Æ dialog bar Æ

Navigator Æ

Hot ListÆ

status bar Æ

← overridewindow

← graphwindow

← ForecastReport window

The Forecast Pro Unlimited window consists of the following seven parts.

Menu bar

Toolbar

Dialog bar

Navigator

Hot List

6 view windows (three are shown above)

Status bar

The Menu bar

The menu provides access to all commands. A complete listing of all menuitems and their associated dialog boxes is presented later in this chapter.

The Toolbar

The icons on the toolbar provide a quick access to the most commonly used commands. Here are the icons and their functions.

New Project . Closes the current forecast project allowing you to start a newone.




Open Project . Opens an existing forecast project.

Save Project . Saves the active forecast project.

Print . Prints the currently selected view.

Copy. Copies the current selection to the Windows clipboard.

Paste. Inserts the contents of the Windows clipboard into the currentinsertion point.

Script . Accesses the Define Script dialog box allowing you to edit thecurrent script.

Read Data. Reads in the data listed on the current script from its primarysource.

Forecast . Executes the current script file and creates the forecasts. Thisoption is only available when a script has been defined and the data have

been read.

View Forecast Report . Opens and closes the Forecast Report view.

View Graph. Opens and closes the Graph view.

Override Forecasts. Opens and closes the Override view allowing you toadjust the forecasts for the current selection.

View Override Report . Opens and closes the Override Report view.

View Outlier Report . Opens and closes the Outlier Report view.

Preview Numeric Output . Opens and closes the Preview Numeric Outputview. This view previews the contents and format of the Numeric Outputfile.

Save Formatted Forecast Report . Allows you to save Formatted ForecastReports to disk. The Excel file will contain a report for every itemforecasted. You can save Formatted Forecast Reports for the items on thecurrent Hot List only using the Hot List’s context menu.



The Forecast Pro Unlimi ted Interface 137

Save Numeric Output . Allows you to save a Numeric Output file to disk. Thefile will contain output for every item forecasted. You can save a NumericOutput file for the items on the current Hot List only using the Hot List’scontext menu.

Options. Opens the Options dialog box allowing you to change the settingsfor the current forecast project.

The Dialog bar

The dialog bar is used to display the current script line, to set the span of thedata and to define a holdout sample.

Defining the fit set . All time series methods begin by fitting the coefficients of a model to historic data. You can alter the beginning year and beginning period to specify the first point of the fitting sample and/or the ending year and ending period to specify the last point.

The dialog bar automatically selects the largest fit set for which data exists. Normally, you will want to accept the defaults. However, you may want toignore earlier data if you distrust their relevance.

Defining a holdout sample. Edit the holdout sample box to specify a number of data points to be withheld from the end of your data set. If you withhold 2or more points, Forecast Pro Unlimited automatically generates out-of-sampleevaluation statistics for any models that you build. If you do not want to perform out-of-sample testing, set the holdout sample to zero.

The Navigator

The Navigator is the primary way to select an item to view in the forecastreport, graph and override windows. After the data have been read, the Navigator displays the available time series in a tree structure. Selecting anitem on the Navigator will automatically display all relevant information inthe open views.

The Navigator also features color-coded icons to allow you to spot items thatcontain overrides and/or comments. A red icon indicates that the item contains




an override and/or comment. A green icon indicates that the item does notcontain an override or comment. A yellow icon on a group indicates that atleast one item further down that branch of the tree contains an override and/or comment.

The Navigator’s context menu provides a convenient way to build a Hot List,apply forecast modifiers and control the display of the Navigator tree. Most of the options are self explanatory, however a few warrant some explanation.

The options under Add to Hot List are used to copy items onto the currentHot List (the Hot List is described in the next section). Children, Parents and Siblings refer to one level down on the current Navigator branch, one level upon the current branch and the same level on the current branch respectively.On the Analysis submenu, Overrides refer to items with direct overrides and

Affected refers to items with indirect overrides (i.e., items where a directoverride elsewhere in the hierarchy changed the item’s forecast).

The Model , Events , Weights , Outliers , Top down and Indexes options areall used to specify forecast modifiers. The most commonly used modifiers can be applied directly using the menu options. The less commonly used

modifiers can be entered from the keyboard using the Custom Modifier

option. A list of all supported modifiers and their function is found in theUsing Forecast Modifiers chapter of this manual.




and are independent of the current Navigator selection). The global views allhave yellow icons.

A description of each of the views appears below.

The Status bar

As you use Forecast Pro Unlimited the status bar displays relevantinformation.

The Forecast Report View

The Forecast Report view is a scrollable text display containing informationabout the forecasts and how they were generated. In addition to providing a

convenient way to view information while using Forecast Pro Unlimited, theForecast Report view is also used to design the text section of the formatted forecast reports that are output to Excel.

The Forecast Report Design dialog box allows you to customize the Forecast

Report view. This dialog box is invoked by selecting Settings>Forecast

Report Design or by selecting Forecast Report Design from the ForecastReport view’s context menu.

You can either select one of the four standardized report styles or define acustom report. Include graph in output can be used in conjunction with anystyle report and specifies whether to include or omit a graph when saving aformatted forecast report to Excel. A description of each report style appears below.

Forecast Only is a minimal display consisting solely of the forecasts.



The Graph View 141

Basic displays a listing of the forecasting model used, a set of the mostcommonly used within-sample statistics, and a forecast display includingconfidence limits and summary statistics.

Standard is the default display. It includes the expert selection logic (if applicable), the model details including model coefficients, a full set of within-sample statistics, a listing of detected/corrected outliers (if outlier

detection is active), a listing of any overrides and the same forecast displayfound in the Basic style.

Complete displays everything found in the Standard display as well as anumeric listing of the historic values, fitted values, converted forecasts, safetystocks and an expanded forecast display.

Custom allows you complete control over what is included in the display.Your selections under the Custom option are automatically retained for the project allowing you to switch between the Custom style and one or more of the standard styles without having to reset your custom settings.

Important note: The Custom option allows you to set the numeric precisionfor the within-sample statistics. These precision settings are used for bothCustom and standard report styles.

The Graph View

The graph view is used to display variables and forecasts graphically. Inaddition to providing a convenient way to view the forecasts while usingForecast Pro Unlimited, the graph can also be included in the formatted forecast reports that you save to Excel. The format and content of theformatted forecast report’s graph will match the current settings for the graphview.




The Graph Settings dialog box allows you to customize the Graph view. This

dialog box is invoked by selecting Settings>Graph Settings or by selecting

Graph Settings from the Graph view’s context menu.

The Settings dialog box contains four pages or tabs. We will discuss each inturn.

The Include tab allows you to the select components to be displayed on thegraph. The components can also be toggled on or off using the Graph view’scontext menu.

Selecting Data only toggles off all of the active components with theexception of the history.



The Graph View 143

The Components tab includes three sections.

The Display section allows you to select the graph type.

Time series displays a graph where the y-axis covers both the historic and forecast period. The All option will display the entire data set in anonscrollable display. If the All option is not selected, the Periods optionallows you to set how many periods should be included in a scrollable display.

Year over year displays a graph where the y-axis is one year long and the data

for each year are “stacked” on the display. The Years option allows you tospecify the number of years to include in the display.

Important note: Because the two graph types are fundamentally different,Forecast Pro Unlimited maintains your settings for these graph typesseparately.

The Labels section allows you to define labels for the X and Y axis and add titles to the graph. The include variable name option will use the variablename followed by the description as the graph’s subtitle. The Clear buttonerases the current labels.

The Y-Axis Scale section controls the scaling.

Automatic allows Forecast Pro Unlimited to select the Y-axis scale.

Custom lets you set the minimum and maximum for the Y-axis scale.

Include zero begins the y-axis at zero or at the minimum negative value.

Relative scaling is only relevant when you are graphing more than onevariable. If it is on, each variable will be displayed on its own scale. If it is




off, all variables will be displayed on the same scale. It is important to knowwhether relative scaling is on or off when you interpret the graph.

Most of the options in the Layout tab are self-explanatory. The Style sectionallows you to set the graph style. The Line Colors and Widths section allowsyou to specify the line colors and widths for the different variables and forecast components displayed. The Include section allows you to toggle on or off various components of the graph. If the Show Ancestry option is selected and the variable name is displayed on the graph, the variable name will appear preceded by any applicable parent group names.

The Items tab allows you to choose up to five variables you want to graph.

The Override Forecasts View

The override view is used to enter overrides to the statistically generated forecasts. Changes made at any level of the forecasting hierarchy willautomatically reconcile all levels. Refer to the Reconciliation Modifiers section of the Using Forecast Modifiers chapter for a discussion on how thereconciliation is accomplished.



The Override Forecasts View 145

Up to ten override rows can be used and you can adjust the forecasts using percentages, increments or by typing in new values. You can also enter comments for any cell on the display.

There is an Overrides tab on the Settings>Options dialog box that allows youto control the document-level options, including how overrides should beapplied (replacement values vs. incremental adjustments) and the precision of

the display. This tab is fully documented in the Menu Items and Dialog Boxes section of this Command Reference chapter.

There are four rows in the example shown above.

Statistical contains the statistically based forecasts generated by Forecast ProUnlimited prior to any overrides being applied.

Override 1 and Override 2 contain any overrides that you have entered for thecurrent forecasts. Proposed overrides are displayed in red, committed overrides are displayed in black. The Commit button is used to convert proposed overrides to committed overrides. The Formula checkbox allowsyou to enter overrides as either formulas or “hard numbers”. The overrideview’s context menu allows you to rename the override row labels and add or

subtract rows on the fly.

Forecast contains the current committed forecasts. These are the values thatare displayed on the graph, and the values that will be written as “forecasts” in



The Override Repor t View 147

The Override Report View

The Override Report view lists items where overrides and/or comments have

been made. To open the Override Report view, select View>Override Report

or click the yellow View Override Report icon ( ).

If you double click an item on the Override Report, the Navigator will jumpdirectly to that item. Thus, you can use the Override Report as the equivalentof a Hot List to navigate through the listed items. All of the global reportviews (yellow icon views) support this kind of navigation.

The content and format of the report is controlled using the Override Reportview’s context menu. The context menu also allows you to print the currentlydisplayed report and save it to Excel.

The Outl ier Report View

The outlier report view lists items where outliers have been detected and/or corrected. To open the outlier report view, select View>Outlier Report or click the yellow View Outlier Report icon ( ).

The content and format of the report is controlled using the outlier report’scontext menu. The context menu also allows you to print the currentlydisplayed report and save it to Excel.

The Outliers tab of the Settings>Options dialog box includes an Include

iteration statistics in reports option. If this option is active the outlier reportwill include detailed statistics describing the outlier detection. Consult the

Forecast Pro Statistical Reference Manual for details.

The Numeric Output View

The Numeric Output view displays the contents and format of the currentlyspecified numeric output file. When designing the numeric output, it is usefulto have the Numeric Output view open.




Selecting Settings>Numeric Output Design opens the Numeric OutputDesign dialog box which is used to specify the format and content of the Numeric Output file (and the Numeric Output view).

The Numeric Output Design dialog box contains both a Set as Default buttonand an OK button. Clicking the OK button will save your current settings for use with the current project. Clicking the Set as Default button will save your

current settings for use with the current project and also save them as thedefault settings for all new projects.

The Numeric Output Design dialog box contains up to three pages or tabs.(The Statistical Output tab is only visible when the “Time series and statistics” option is chosen on the Output Format tab.) We will discuss eachtab in turn.

The Output Format tab is used to specify the basic layout for the NumericOutput file.

Time series only vs. Time series and statistics. Selecting Time series and

statistics displays the Statistical Output tab allowing you to include within-sample statistics and/or information on the forecasting model used in the Numeric Output file.

Forecast file prefix. Specify the character(s) to use as the default prefix whennaming the Numeric Output file. If the Single Output file option is selected in



The Numeric Output View 149

the Layout section, the default Numeric Output filename will be the prefixfollowed by the project name followed by “Numeric Output”. If the Single

Output file option is not selected, each output filename will be the prefixfollowed by the corresponding input filename.

The Layout section of this tab allows you to specify the default data mode, theorientation (row vs. column), whether or not to include group level output and

whether or not to include item-level data.

Single output file. Output a single Numeric Output file containing informationfor all forecasted items will be written. If Single output file is not selected,then a separate Numeric Output file will be written for each input file.

Break record up into lines. If selected, then each Numeric Output componentspecified on the Forecast Output tab (see below) will appear on a separate lineof the Numeric Output file.

The Text Format section of this tab allows you to specify the conventions to

use when outputting a text file. You will only be able to edit these settingswhen the data mode is set to Txt (text) or Mlt (text input file format).

The Forecast Output tab is used to define the data label fields, to select theforecast components to include and to set their precision. If you are outputtinga fixed width text file then you may also set the column widths.

Attributes. Include separate attribute fields for each level of the hierarchy. If you used attribute fields in your input data files to define groups then you willmost likely want to select this option so that your output file will match your input file. Consult the Setting Up Your Data chapter for a discussion of attribute fields.



Menu Items and Dialog Boxes 151

Menu Items and Dialog Boxes

This section covers all of the Forecast Pro Unlimited menu items and their associated dialog boxes.

The File Menu

The file menu is used to manipulate forecast project files, save ODBCconnections strings for use in command-line operation, print the currentlydisplayed forecast report and to exit Forecast Pro Unlimited.

File>New is used to open a new forecast project. This task can also beaccomplished using the blue New Project icon ( ).

File>Open is used to open an existing forecast project. This task can also beaccomplished using the blue Open Project icon ( ).

File>Close is used to close the current forecast project.

File>Save is used to save the current forecast project using the currentlyspecified name. This task can also be accomplished using the blue SaveProject icon ( ).

File>Save as is used to save and name the current forecast project.

File>Print is used to print the currently selected view. This task can also beaccomplished using the blue Print icon ( ).

File>Print Preview is used to display the currently selected view formatted as it will be printed (see File>Print above). Checking the appearance of theoutput prior to printing can save you trips to the printer.

File>Print Setup is used to change various printing options.

File>Exit is used to exit Forecast Pro Unlimited.




The Settings Menu

The settings menu provides access to the Options dialog box which allowsyou to change your project settings and to other dialog boxes which allow youto customize the various Forecast Pro Unlimited views.

With the exception of Settings>Options which is described below, all of the

other menu options lead to dialog boxes that are described in detail in theForecast Pro Unlimited Interface chapter.

Settings>Options

Selecting Settings>Options opens the Options dialog box which is used tochange various Forecast Pro Unlimited settings.

The Settings dialog box contains both a Set as Default button and an OK button. Clicking the OK button will save your current settings for use with thecurrent project. Clicking the Set as Default button will save your currentsettings for use with the current project and also save them as the defaultsettings for all new projects.

The Settings dialog box contains six pages or tabs. We will discuss each inturn.

The Basic tab contains frequently accessed settings that every user should understand.

Input Data. Specify the complete path name of the directory where the program should read data.




Projects. Specify the complete path name of the directory where the programshould write the project files.

Output. Specify the complete path name of the directory where the programshould write all output files (e.g., numeric output files, forecast reports, etc.).

Data mode. Select the type of data file that you will be using from the list of

supported types.

Allow negative forecasts. Most business data are inherently positive, althoughthere are exceptions like telephone connection gain or sales data that includereturns as negatives. Users with positive data are understandably disturbed when their forecasts turn negative. This can occur when Forecast ProUnlimited captures a downward trend at the end of your historic data. If Allow

negative forecasts is turned off, Forecast Pro Unlimited will clip forecasts tozero that would otherwise be negative. If Allow negative forecasts is turned on, then Forecast Pro Unlimited takes no special actions to avoid negativeforecasts.

Forecast horizon. Specify how many periods ahead you want Forecast ProUnlimited to forecast.

Upper confidence limit. Specify the percentile for the upper confidence limits.The upper confidence limit for a specific item can also be specified using the\Upper= modifier.

Lower confidence limit. Specify the percentile for lower confidence limits.The lower confidence limit for a specific item can also be specified using the\Lower= modifier.

Safety stock lead time. Specify the lead time for the safety stocks. Thespecified lead time appears highlighted in the Safety Stock section of theForecast Report and determines the values written to the numeric output file.




The Formats tab controls the formatting of the input and output.

Decimal separator. Specify whether a period or a comma is used for thedecimal point. The program’s default is the convention appropriate for theUnited States, which is the period.

Column delimiter. Specify whether items in an MLT file are separated byspaces, commas or tabs (i.e., are the files space delimited, comma delimited or tab delimited).

Thousands separator. Specify the thousands separator used for input MLTfiles. The choices are none, comma and period. Forecast Pro Unlimited doesnot use thousand separators in output.

Ignore leading zeros. If this option is selected then leading zeros prior to thefirst nonzero data point will be ignored (i.e., the data has not started yet and the zeros are interpreted as placeholders). For a more complete discussion onthe handling of missing data and the distinction between a zero value and amissing value, please refer to the Missing Values and Zeros section in the

Setting Up Your Data chapter.

Ignore trailing zeros. If this option is selected then trailing zeros after the lastnonzero data point will be ignored (i.e., the data has ended and the zeros areinterpreted as placeholders). This means that the time series would beconsidered “dead” in multiple-level scripts and not forecasted. In nonmultiple-level scripts the forecasts for the time series would begin after the last nonzero point. For a more complete discussion on the handling of missing data and thedistinction between a zero value and a missing value, please refer to theMissing Values and Zeros section in the Setting Up Your Data chapter.




Missing Values. Specify how you wish the program to treat missing values.The options are: truncate the data set (i.e. discard all data that precedes themissing value); impute the missing data or set missing data points to zero.Regardless of how this option is set, missing values that appear prior to thefirst data point are ignored. For a more complete discussion on the handling of missing data and the distinction between a zero value and a missing value, please refer to the Missing Values and Zeros section in the chapter entitled

Setting Up Your Data.

Include group totals for data files. If this switch is on, Forecast Pro Unlimited will automatically include a group total for each data file listed on the script.Including or not including a group total for the data files will yield differenthierarchies on the Navigator.

Text file encoding. Specify the format to use when writing two-byte charactersto text files. This option is only of relevance if your data or scripts utilize non-ASCII characters. (Non-Latin based languages such as Japanese, Chinese,

Russian, Arabic, etc. use non-ASCII characters.)

Code page. Specify the Windows code page to use when text file encoding isset to multi-byte. The code page specifies the character set to use whenwriting a multi-byte text file. (Most code pages are designed to support aspecific language.)

Display Precision: Numeric. Specify the precision (i.e., number of decimal places) to use when displaying time series data in the view windows.

Display Precision: Percentage. Specify the precision (i.e., number of decimal places) to use when displaying percentages in the view windows.




The Outliers tab is used to specify whether or not to detect and/or correctoutliers. It also allows you to set the sensitivity settings for outlier detectionand to dictate the amount of detail in the outlier report.

None turns outlier detection off. The forecasts will be generated using theuncorrected history.

Detection only will detect outliers and display the suggested corrected values,however, the forecasts will be generated using the uncorrected history.

Detection and correction will detect outliers and will automatically use thecorrected values when generating forecasts.

Sensitivity (std deviations) allows you to set the sensitivity of the outlier detection algorithm. If a given fitted error exceeds this threshold and it is the

largest error detected during the current iteration it will be flagged as anoutlier. Consult the Forecast Pro Statistical Reference Manual for details onhow outlier detection and correction works.

Maximum iterations allows you to set the maximum number of iterations permitted during outlier detection for a given item. This setting thereby alsodefines the maximum number of outliers than can be detected for a givenitem. Consult the Forecast Pro Statistical Reference Manual for details onhow outlier detection and correction works.

Include iteration statistics in reports allows you to include detailed statisticsdescribing the outlier detection process in the outlier report. Consult theForecast Pro Statistical Reference Manual for details.

The Overrides tab controls operation and display of the Override view.




Adjustment mode. If this selection is set to “Override” any overrides entered will replace the corresponding statistical forecasts (or overrides entered on a preceding row). If this switch is set to “Incremental” overrides you enter will be added as incremental adjustments to the statistical forecast rather thanreplacing it. Thus the final forecast will equal the sum of the statisticalforecast and all overrides entered. If you attempt to change this setting whenoverrides exist for the current project, you will receive a warning and all

overrides will be erased.

Manual reconciliation mode. If this switch is on, the override window willinclude a Retain button as well as a Commit button. The Retain button is used to accept overrides without reconciling the hierarchy. This is useful whenworking with large complex hierarchies where the reconciliation process takessome time.

In manual reconciliation mode you would typically enter and retain all of theoverrides you wish to make (i.e., for multiple items) and then click theCommit button to perform the reconciliation once, rather than having to wait

for the program to reconcile after each item that you override.

Number of adjustment rows. This setting controls the number of adjustmentrows displayed in the Override view.

Allow disaggregation. If this switch is on, the Override window will include a“Disaggregate to” drop down box, allowing you to disaggregate (i.e., pushdown) any overrides made at a group level to a lower-level of the hierarchy.The disaggregation is based on a proportional allocation.

Forecast Pro Unlimited supports command-line operation allowing you tocreate forecasts in a “hands-off” mode. For a complete description of how torun the product from the command line consult the Operating From the




Command Line lesson in the tutorial and the Command-line Operation sectionof this chapter.

The Command Line tab allows you to dictate the options to apply if thecurrent project is run via the command line.

The first three options control how the forecasts are generated when the

project is run from the command line.

Modifiers. If this option is selected, the project’s forecast modifiers (if any are present) will be used to generate the forecasts in command-line mode. If thisoption is not selected, all forecast modifiers in the project will be ignored when running from the command line.

Overrides. If this option is selected, the command-line run will begin by

reading in the data, generating the statistical forecasts and then it will applyany matching overrides/comments in the project to the new forecasts. If thisoption is not selected, all overrides in the project will be ignored when

running from the command line.

Settings. If this option is selected, the command-line run will use all of thesettings associated with the project (e.g., all settings in the Settings dialog box, numeric output formatting, etc.). If this option is not selected, the defaultsettings (i.e., the settings that have been saved as the defaults for use with new projects) will be used when running from the command line. Project settingsare saved in the project file (*.FPProj). Default settings are stored in theForecastProUnlimited.ini file.

The remaining options specify the output files to generate when the project isrun from the command line.

Numeric output. If this option is selected, the command-line run will save thenumeric output. You have the option of saving this file for all items forecasted or (if you select Hot List only) just for the items listed on the Hot List.

Forecast Report. If this option is selected, the command-line run will save aformatted forecast report file to Excel. You have the option of saving this filefor all items forecasted or (if you select Hot List only) just for the items listed on the Hot List. Keep in mind that these files can potentially be very large.




Outlier report and Override report. If either or both of these options areselected, the command-line run will save the specified report(s). The formatand contents (including all items vs. Hot List only) will match the currentsettings in the project (or the defaults if you’ve specified to use them).

Project (FPProj). If this option is selected, the command-line run will updateand save new project files. Normally the only part of the project that will

change during the command-line run is the binary file (*.fcb). This is a“snapshot” file that allows you to open up the project without reading in thedata and generating new forecasts.

The Advanced Controls tab allows you to set expert selection options, controlthe timing of some program operations and adjust how frequently ForecastPro Unlimited checks for program updates. Adjusting the expert selection and performance options can improve processing speed for forecasting, applyingforecast modifiers and saving projects.

Exclude ARIMA. If this switch is on Forecast Pro Unlimited’s expert selectionmode will not consider Box-Jenkins models. This will speed the forecastgeneration process considerably.

Exclude seasonal simplification. If this switch is on Forecast Pro Unlimited’sexpert selection mode will not consider seasonally simplified forms of exponential smoothing models. If you are forecasting data where the number of periods per cycle is greater than 13 (e.g., weekly data) this will speed theforecast generation process considerably.

Automatic recalculation. If this option is selected, Forecast Pro Unlimited willautomatically recalculate the forecast and re-reconcile the hierarchy everytime you apply a forecast modifier. For large hierarchies this can be time




consuming. If you turn this option off, when you specify a forecast modifier the forecasts will not be updated until you click the red Forecast icon. Thisallows you to specify multiple forecast modifiers without having to wait for the program to recalculate the forecasts in between each specification.

Display dates generically. If this option is selected then monthly labels will bedisplayed using period numbers rather than the names of the months. For

example, “2007-01” will be used rather than “2007-Jan”. This option is oftenused when working with a fiscal calendar where displaying the names of themonths could be confusing.

Check for product updates automatically. If this option is selected, ForecastPro Unlimited will periodically check online for product updates and alert youif they are available. If your support and maintenance contract is up to date, product updates are free.

The Operations Menu

The operations menu is used to perform basic program operations. Most of theactions can also be performed via icons.

Operations>ODBC Connect

Operations>ODBC Connect is used to connect to an ODBC database. After selecting this option you will be prompted for the type of database and thedatabase file. This option is only available when the data mode is set toODBC.

Operations>Script

Operations>Script accesses the Define Script dialog box allowing you to editthe current script. This dialog box can also be accessed via the red Script icon( ). Please read the chapter entitled Defining the Script for complete details

on the script.




Operations>Read Data

Operations>Read Data reads in the data listed on the current script from its primary source. This task can also be accomplished using the red Read Dataicon ( ).

Operations>Forecast

Operations>Forecast executes the current script file and creates theforecasts. This option is only available when a script has been defined and thedata has been read in. This task can also be accomplished using the red Scripticon ( ).

The Project Menu

The project menu is used to import overrides and modifiers from other projects, to export (i.e., save to disk) numeric output files and report files and

to set project passwords.

Project>Import

Project>Import>Overrides from is used to import overrides and their associated comments from a saved project into the current project. In order for a value to be imported, the item name, ancestry and date all need to match anitem in the current project.

Project>Import>Modifiers from is used to import forecast modifiers from asaved project into the current project. In order for a modifier to be imported

the item name and ancestry must match an item in the current project.

Project>Export

The Project>Export menu allows you to save numeric output files and allForecast Pro Unlimited report files. Consult the appropriate section of theForecast Pro Unlimited Interface chapter for details on designing the contentand format of these output files.




Project>Set password

Project>Set password is used to password protect your project. When youattempt to open a password-protected project, Forecast Pro Unlimited will prompt you for the password and only open the project if the correct password is entered.

When you password protect a project, Forecast Pro Unlimited uses MicrosoftAccess’s security functionality to password protect the project’s .mdb file.This means that in addition to requiring the password to open the project inForecast Pro Unlimited, the password is also required to open the project’s.mdb file in Microsoft Access.

The View Menu

The View menu allows you to view reports, display graphs and customizeyour display.

View>(All available views)

The first six options on the View menu will toggle the selected view windowon and off. This can also be accomplished using the appropriate green or yellow icon.

View>Toolbar

View>Toolbar displays and hides the Toolbar, which includes buttons for some of the most common commands in Forecast Pro Unlimited. A check

mark appears next to the menu item when the Toolbar is displayed.

View>Status Bar

View>Status Bar displays and hides the Status Bar, which describes the progress of the currently running forecast job, or the action to be executed bythe selected menu item or depressed toolbar button. A check mark appearsnext to the menu item when the Status Bar is displayed.




View>Modifiers

View>Modifiers displays the currently selected forecast modifiers on the Navigator.

The Window Menu

The Window menu enables you to arrange your open windows in variousways.

Window>Cascade

Window>Cascade arranges all nonminimized windows in an overlapped fashion.

Window>Tile

Window>Tile arranges all nonminimized windows in a tiled (nonoverlapped)fashion.

The Help Menu

Forecast Pro Unlimited includes extensive on-line help.

Help>Help Topics

Help>Help Topics displays an index of all of the help topics available.

Clicking on an entry will jump to the selected topic.

Help>User Guide (PDF)

Help>User Guide (PDF) accesses a pdf version of the Forecast ProUnlimited User’s Guide.




Help>User Guide (PDF)

Help>Statistical Reference (PDF) accesses a pdf version of the Forecast Pro

Statistical Reference Manual.

Help>Check for Updates

Help>Check for Updates will check online to see if you are running thelatest release of the program. If a product update is available you’ll receive amessage alerting you to this fact with instructions describing how to receivethe update. If your support and maintenance contract is up to date, productupdates are free.

Help>About Forecast Pro Unlimi ted

Help>About Forecast Pro Unlimited displays the Forecast Pro Unlimited version number and copyright notice.

Command-line Operation

You can run Forecast Pro Unlimited noninteractively. When the command line to execute the program includes a project filename (*.FPProj) and the /b parameter, Forecast Pro Unlimited will read in the data, create the forecasts,save all output files and then exit. This feature is particularly useful when you

are integrating forecasting with other software systems and need a “hands-off”approach.

To drive Forecast Pro Unlimited entirely automatically, follow these two

steps.

1.) Prepare a valid project and place it in your project directory. The easiestway to prepare this file is to use Forecast Pro Unlimited interactively to define

all desired project settings, output formats, a script and (optionally) forecast

modifiers and overrides, and then save the project using use File>Save.

Alternatively, you can generate the file externally but this will requireunderstanding the project file and the (optional) corresponding MDB fileformats.



Command-line Operation 165

2.) Drive Forecast Pro Unlimited by issuing the command

“C:\Program Files\Forecast Pro Unlimited v6\ForecastProUnlimited.exe” Test.FPProj /b

where we assume that the program is located in the directory “C:\ProgramFiles\Forecast Pro Unlimited v6” and a project named Test.FPProj is located in your current project directory. The “/b” parameter is used to indicate that

you wish to run in command-line (batch) mode.

You can issue the command in any number of ways, including:

Select Run from the Start menu and issue the command from there.

Set up a shortcut to issue the command.

Issue the command from within a different application using a systemcall.

The Command Line tab of the Settings>Options dialog box allows you to

specify several key command line forecasting options and the output files tosave in command-line mode. This tab’s functionality was fully described inthe Menu Items and Dialog Boxes section earlier in this chapter.

Important Notes

1. If there are spaces in the pathname you must include quotation marksas illustrated in the example above.

2. If you do not want the Forecast Pro Unlimited menu to appear, run the program minimized.



167

Chapter 6

Understanding How Forecast Overridesare Allocated

The override facility in Forecast Pro Unlimited allows you to adjust thestatistically generated forecasts. Changes made at any level of the forecastinghierarchy will automatically reconcile all levels. This chapter was written toclarify exactly how the reconciliation is performed and document cases whereadjustments cannot be fully implemented due to conflicting overrides or program settings.

Adjustments Made to a SingleLevel

In this section we will describe how Forecast Pro Unlimited reconciles theforecast hierarchy when overrides are made to a single level of the hierarchy.

Important: If you restrict your overrides to a single level of the hierarchy, the





168 Understanding How Forecast Overrides are Allocated

Figure 1 below depicts a simple forecast hierarchy prior to any overrides being entered. Notice that at this point the Statistical forecasts equal thecommitted Forecasts.

Total

S = 5

F = 5

GroupA

S = 2

F = 2

GroupB

S = 3

F = 3

ItemA1

S = 1

F = 1

ItemA2

S = 1

F = 1

ItemB1

S = 1

F = 1

ItemB2

S = 2

F = 2

Figure 1.

Let’s now modify the above hierarchy by overriding the forecast for ItemA1

to equal 75. This results in the hierarchy shown in Figure 2 below.

Total

S = 5

F = 79

GroupA

S = 2

F = 76

GroupB

S = 3

F = 3

ItemA1

S = 1

O = 75

F = 75

ItemA2

S = 1

F = 1

ItemB1

S = 1

F = 1

ItemB2

S = 2

F = 2

Figure 2.

Notice that at this point the committed Forecasts for GroupA and Total do notequal the Statistical forecasts.

If we had started with the hierarchy depicted in Figure 1 and adjusted theGroupB forecast to 75 (rather than ItemA1) we would generate Figure 3.



Adjustments Made to More Than One Level 169

Total

S = 5

F = 77

GroupA

S = 2

F = 2

GroupB

S = 3

O = 75

F = 75

ItemA1

S = 1

F = 1

ItemA2

S = 1

F = 1

ItemB1

S = 1

F = 25

ItemB2

S = 2

F = 50

Figure 3.

Notice that the committed Forecasts for ItemB1, ItemB2 and Total do notequal the Statistical forecasts. Notice also that Forecast Pro Unlimited allocated the GroupB override to ItemB1 and ItemB2 based on the proportionsestablished by their Statistical forecasts.

Adjustments Made to More ThanOne Level

In this section we will describe how Forecast Pro Unlimited reconciles theforecast hierarchy when overrides are made to more than one level of thehierarchy.

Important: There are many ways that this type of reconciliation could be

performed—each one leading to different forecasts. If you plan to make

overrides at more than one level of the hierarchy, it is essential that you

understand how Forecast Pro Unlimited performs the reconciliation.

Figure 4 below depicts a forecast hierarchy where ItemA1 has beenoverridden to equal 75. (It is identical to Figure 2.)




Total

S = 5

F = 79

GroupA

S = 2

F = 76

GroupB

S = 3

F = 3

ItemA1 S = 1

O = 75

F = 75

ItemA2S = 1

F = 1

ItemB1S = 1

F = 1

ItemB2S = 2

F = 2

Figure 4.

Let’s now modify the above hierarchy by overriding the forecast for Total toequal 500. This results in the hierarchy shown in Figure 5 below.

Total

S = 5

O = 500

F = 500

GroupA

S = 2

F = 200

GroupB

S = 3

F = 300

ItemA1

S = 1

O = 75

F = 75

ItemA2

S = 1

F = 125

ItemB1

S = 1

F = 100

ItemB2

S = 2

F = 200

Figure 5.

Notice that:

1.) Forecast Pro Unlimited used the Statistical forecasts for GroupA and GroupB to determine the allocation proportions NOT the previouslycommitted Forecasts displayed in Figure 4. Using the committed Forecastsfrom Figure 4 would have resulted in a dramatically different allocation.

2.) Forecast Pro Unlimited did not alter the committed Forecast value for ItemA1 because it had been explicitly overridden. When Forecast Pro




Total

S = 5

F = 302

GroupA

S = 2

F = 150

GroupB

S = 3

F = 152

ItemA1

S = 1

O = 75

F = 75

ItemA2

S = 1

O = 75

F = 75

ItemB1

S = 1

O = 150

F = 150

ItemB2

S = 2

F = 2

Figure 7.

Suppose you try to override the forecast for GroupA to equal 100. Theoverride could not be applied because both ItemA1 and ItemA2 are “locked”.Forecast Pro Unlimited will display an error message and reject the override.

Let’s now modify the above hierarchy by overriding the forecast for GroupB

to equal 50. This results in the hierarchy shown in Figure 8 below.

Total

S = 5

F = 200

GroupA

S = 2

F = 150

GroupB

S = 3

O = 50

F = 50

ItemA1

S = 1O = 75

F = 75

ItemA2

S = 1O = 75

F = 75

ItemB1

S = 1O = 150

F = 150

ItemB2

S = 2F = -100

Figure 8.

Notice that the override for GroupB resulted in a negative committed Forecastfor ItemB2. This result assumes that the “Allow Negative Forecasts” setting in

the Forecast Pro Unlimited Options dialog box was selected. If this settingwas not selected, the override could not be made and Forecast Pro Unlimited would display an error message and reject the override.



Summation 173

Summation

It should be quite apparent that if you make adjustments to more than onelevel of a hierarchy the reconciliation becomes quite complex. Althoughseveral of the examples presented seemed involved, they were actually quitesimple hierarchies consisting of only 3 levels and 4 end-items. Most users of

Forecast Pro Unlimited are working with thousands of items and much morecomplex hierarchies.

In developing the reconciliation routine for Forecast Pro Unlimited theauthors considered using several different algorithms. In many situations thechoice of which algorithm to use would change the forecast values— sometimes significantly. None of the algorithms were “statistically superior”to the others—they were simply different.

Virtually all of the complexity (and differences between algorithms)disappears if you restrict yourself to making forecast adjustments to a singlelevel of the hierarchy. Although we’ve stated this previously, it’s worthrepeating:

If you restrict your overrides to a single level of the hierarchy, the





175

Appendixes

Speeding Up the Program

If decreasing processing time is important for your application, you may wishto investigate some of the following options.

1. Hardware

Upgrading your hardware will improve performance dramatically. ForecastPro Unlimited will run on any Intel processor from the 386 on. Since this product deals with large amounts of floating point data, it makes sense to runit on as fast a machine as possible.

As an illustration, using expert selection and four years of monthly history per item, a Core2/2.2GHz machine forecasted 10,000 items in 39 seconds. Thesame run took 4 minutes on a Pentium III/600MHz machine and 24 minuteson a Pentium/90MHz.

Having enough RAM is also important—we recommend 2 Gigs. As you usethe program, Forecast Pro Unlimited is storing and retrieving a great deal of information in memory. If you do not have an adequate amount of RAMForecast Pro Unlimited will resort to using virtual memory (i.e., caching to



176 Appendixes

your hard drive). This is substantially slower than using RAM. Increasing theamount of RAM in your computer will improve performance of all of your applications.

2. Manual Override Mode

If you make overrides to large complex hierarchies, the reconciliation can take

a little time. To minimize the reconciliation time, Forecast Pro Unlimited supports a manual override mode. When this mode is active, the overridewindow will include a Retain button as well as a Commit Button. The Retain button is used to accept overrides without reconciling the hierarchy. In thismode you would typically enter and retain all of the overrides you wish tomake (i.e., for multiple items) and then click Commit to perform thereconciliation. The control to turn on manual override mode is found on the

Overrides tab of the Settings>Options dialog box.

3. Manual Recalculation

By default, Forecast Pro Unlimited will automatically recalculate the forecastsand re-reconcile the hierarchy every time you apply a forecast modifier. For large hierarchies this can be time consuming. Forecast Pro Unlimited allowsyou turn off the automatic recalculation using an option of the Advanced

Controls tab of the Settings>Options menu. If you turn this option off, whenyou specify a forecast modifier the forecasts will not be updated until youclick the red Forecast icon ( ). This allows you to specify multiple forecast

modifiers without having to wait for the program to recalculate the forecastsin between each specification.

4. Outlier Detection and Correction

The outlier detection and correction algorithm will slow down the processing.If you wish to use outlier detection and correction but speed is of concern, oneoption is to initially run the forecasts in the detection only mode to detect theoutliers and then turn outlier detection off and use the \OUTLIER=CORRECTmodifier to perform the corrections on the detected items.

5. Data Files

If you are using xlsx files you might consider switching to xls files. Xlsx filesare slower to read than xls files and use twice as much memory. Also if you



End-User License Agreement 177

are using large spreadsheet files breaking them up into several smaller files or increasing the amount of RAM available will speed up processing.

6. Model Selection

Specifying “Exclude ARIMA” on the Advanced Controls tab of the

Settings>Options dialog box will speed up forecast generation considerably.

This option will remove Box-Jenkins models from consideration when usingexpert selection.

If you are forecasting data where the number of periods per cycle is greater than 13 (e.g., weekly data) you might consider turning on the “Excludeseasonal simplification” option in the Advanced Controls tab of the

Settings>Options dialog box. If this switch is on, Forecast Pro Unlimited’sexpert selection mode will not consider seasonally simplified forms of exponential smoothing models. This will speed up forecast generationconsiderably

Specifying the models to use rather than using the expert selection willsubstantially improve performance. This will require some work to determinethe best type of model to use for your data. The best way to determine theappropriate model for your data set is to experiment using the forecastevaluation procedures (see methodology chapter in the Forecast Pro

Statistical Reference Manual for details).

End-User License Agreement

This End-User License Agreement (“EULA”) is a legal agreement between

Licensee (either an individual or a single entity) and Business ForecastSystems, Inc. (“BFS”) for the BFS software that accompanies this EULA.

1. DEFINITIONS

A. “BFS” means Business Forecast Systems, Inc., 68 Leonard Street,Belmont, MA 02478 USA, the author and owner of the copyright of thiscomputer program product.



178 Appendixes

B. “Computer” means the single computer, workstation, terminal or other digital electronic device on which You use this program.

C. “Software” means the set of computer software programs in this package,regardless of the form in which You subsequently use it.

D. “Licensee” means an individual or single entity exercising rights under and

complying with all the terms of this EULA. In the remainder of this EULA,the Licensee is referred to as “You.”

E. “Authorized User” means the Licensee in the case where the Licensee is anindividual. In the case where the Licensee is an entity, “Authorized User”means (i) an employee of the Licensee, but only while such employee isacting within the scope of such employee’s employment by Licensee; and (ii)the primary user of the Computer on which the Software is installed.

If You do not agree with the terms of this EULA, BFS does not grant anylicense to the Software and You should not download, install or use the

Software. In such event that You do not agree with the terms of this EULAand have already installed or downloaded the Software, You must remove theSoftware from your Computer and return to BFS the complete software package, including disks and printed materials, and destroy all copies (or, if You downloaded the Software, notify BFS in writing that You have removed the Software from your Computer and destroyed all copies). If You do notcomply with these return and notification requirements within thirty days (30)of your receipt of the Software, You will not be entitled to a refund of anylicense fees You paid.

2. OWNERSHIP

BFS retains title, ownership, and all rights and interests in and to theSoftware, documentation and all other materials supplied by BFS. BFS doesnot sell the Software or any copies thereof, but only grants limited licenses touse the Software in accordance with the terms of this EULA.

3. GRANT OF LICENSE

Upon acceptance of this EULA, BFS grants You a non-transferable, non-exclusive, limited right to use this copy of the Software. You may use theSoftware and the associated documentation only in connection with one



180 Appendixes

5. TRANSFER OR REPRODUCTION

You may not transfer or reproduce the Software with the exception of makinga maximum of two (2) copies of the Software to be used for backup purposesonly. When using a Computer with a hard disk, one copy of the Software may be installed on the hard disk provided that the copy is removed if the Softwareis used on another Computer.

6. LIMITED WARRANTY

THE SOFTWARE, THE DOCUMENTATION, AND ANY OTHER MATERIALS THAT MAY BY BE PROVIDED BY BFS ARE PROVIDED"AS IS" WITHOUT ANY WARRANTY OF ANY KIND. BFS HEREBYEXPRESSLY DISCLAIMS ALL WARRANTIES WITH REGARD TO THESOFTWARE, THE DOCUMENTATION, AND ANY OTHER MATERIALS, WHETHER EXPRESS, IMPLIED, STATUTORY OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY IMPLIEDWARRANTIES OF MERCHANTABILITY, FITNESS FOR A

PARTICULAR PURPOSE, AND ANY WARRANTIES ARISING OUT OFCOURSE OF DEALING OR COURSE OF PERFORMANCE. BFSEXPLICITLY DOES NOT WARRANT THAT THE SOFTWARE WILLMEET ALL OF YOUR REQUIREMENTS OR THAT THE OPERATIONOF THE SOFTWARE WILL BE ERROR-FREE. YOU ACKNOWLEDGETHAT YOU HAVE RELIED ON NO WARRANTIES AND NOWARRANTIES ARE MADE HEREIN BY BFS.

7. LIMITATION OF LIABILITY

BFS SHALL NOT BE LIABLE FOR ANY LOSS OR DAMAGE THAT

MAY BE RELATED TO THE FURNISHING, PERFORMANCE, OR USEOF THE SOFTWARE, INCLUDING WITHOUT LIMITATION ANYINDIRECT, SPECIAL, INCIDENTAL, ECONOMIC, LOST PROFIT, OR CONSEQUENTIAL LOSS OR DAMAGE OF ANY KIND WHATSOEVER EVEN IF BFS HAS BEEN ADVISED OF THE POSSIBILITY OF THOSEDAMAGES. IN NO EVENT SHALL BFS’S LIABILITY UNDER THISEULA EXCEED THE AMOUNT PAID FOR THE SOFTWARE LICENSE,IF ANY.

8. TERM AND TERMINATION



Suggested Reading 181

This EULA is effective until terminated. Your rights under this EULA willterminate automatically without notice from BFS if You fail to comply withany provision of this EULA. Upon termination of this EULA, You mustdestroy all copies, full or partial, of the Software. You agree that upontermination of this EULA for any reason, BFS may take actions so thatSoftware no longer operates.

9. GENERAL TERMS AND CONDITIONS

A. You acknowledge that You have read this EULA understand it, and agreeto be bound by its terms and conditions. This EULA supersedes all prior agreements, oral or written communications, proposals, representations and warranties and prevails over any conflicting or additional terms of any quote,order, acknowledgment, or other communication between the parties relatingto its subject matter during the term of this EULA. No modification of thisEULA will be binding, unless made in writing and signed by an authorized representative of each party.

B. In the event that any portion of this EULA is held unenforceable, theunenforceable portion shall be construed in accordance with applicable law asnearly as possible to reflect the original intentions of the parties, and theremainder of this EULA shall remain in full force and effect.

C. This EULA will be governed by and construed and enforced according tothe laws of the Commonwealth of Massachusetts, USA excluding its conflictof law provisions.

Suggested Reading

The Forecast Pro Statistical Reference Manual is a pdf document that iscopied to the program directory when you install Forecast Pro. It is alsoaccessible via the Forecast Pro help system and is the primary suggested reference for all of the forecasting techniques, statistics and algorithms found in Forecast Pro.

The Forecast Pro Statistical Reference Manual includes an extensive bibliography. A few recommended textbooks include:



182 Appendixes

J. S. Armstrong [2001] Principles of Forecasting: A Handbook for Researchers and Practitioners, Norwell MA: Kluwer AcademicPublishers.

S. Makridakis, S. C. Wheelwright and R.J. Hyndman [1998]Forecasting Methods and Applications, Third Edition, New York:Wiley.

P. Newbold and T. Bos [1990] Introductory Business Forecasting,Cincinnati: South-Western.



183

Index

\ARIMA, 125\BJ, 125\CFIT, 127\CFIT=, 127

EXPO, 127GROW, 127LINE, 127QUAD, 127

\CONST, 125\DISCRETE, 125\EVENT, 53, 128, 129

\EXSM, 124\EXSM=, 124, 128\FIXED=, 127\HOLT, 124\INDEXES, 64, 118, 120, 131\INTER, 126\OUTLIER=, 130\POW, 132\SALY, 126\SALYDI, 127

\SALYDP, 127\SALYI=, 126\SALYP=, 126\SIMPLE, 124\SMA=, 126\TOPDOWN, 63, 65, 118, 119, 120,131\WINTERS, 124\XY, 125, 128 _GROUP, 1171-2-3, 95, 96

Adjustment mode, 157Allow disaggregation, 157Allow negative forecasts, 153Automatic recalculation, 159Bottom-up, 62Box-Jenkins, 14

modifiers, 125Cannibalization, 60, 66Cascade, 163Check for updates, 164



Index 185

M-Competitions, 79Mdb files, 111Missing values, 93, 155MLT, 96, 101Modifiers, 130, 163Multiple-level forecasting, 116 Navigator, 137

Negative forecasts, 153 New, 151 Numeric Output Design, 148 Numeric Output Files, 111 Numeric Output View, 147 Numeric Precision, 155ODBC, 96, 103ODBC Connect, 160Open, 151Operations Forecast, 161Operations ODBC Connect, 160

Operations Read Data, 161Operations Script, 160Options dialog box, 152Outlier detection/correction, 129, 156Outlier Report view, 147Outlier reports, 113Outliers, 59Output directory, 153Override Report view, 147Override reports, 113Override view, 144Passwords, 162Percentage precision, 155Periods per cycle, 95Periods per year, 95Point forecast, 12Power transformation, 132Print Preview, 151Print Setup, 151Printing, 151Promotions, 52, 60

Read Data, 161Safety stock lead time, 153Save, 151Save as, 151Script, 115, 160

defining, 115multiple-level, 116

overview, 21Seasonal indexes, 57Sensitivity outlier setting, 156Settings Options, 152Short lived products, 66Simple moving average, 13Speeding up the program, 175Spreadsheet data, 95, 96Starting Forecast Pro Unlimited, 19Starting period, 94Starting year, 94

Statistical row, 145Status bar, 162System requirements, 7Text file encoding, 155Thousands separator, 154Tile, 163Toolbar, 135, 162Top-down, 63Trailing zeros, 154Updates, 164Variable

description, 94header information, 94name, 94

Very simple models, 126View Modifiers, 163View Status bar, 162View Toolbar, 162Weekly data, 55Weighting transformation, 129Window Cascade, 163



Window Tile, 163Winters, 57WK*, 95, 96XLS, 95, 96Zeros, 93

Forecast Pro Unlimited V6 User's Guide

Documents

Transcript of Forecast Pro Unlimited V6 User's Guide